1. I'trying to do a firstime kernel build with netem and cannot seem
to find thproper build option in menuconfig (with both 2.4.28 and
2.6.8.1 kernels). Aearlier posting claims thbuild option can be
found under thesmenu options:
Networking options -->
QoS and/or fair queueing
Network Emulator
However thmenuconfig options for both thaforementioned kernel revs I
seare:
Networking options -->
QoS and/or fair queuing
Network Testing -->
PackeGenerator
Is thPackeGenerator option in fact 'netem'?
2. Itrying to selecthe proper version of iproute to download, it
appears theris only onversion available for 2.4.x kernels:
iproute2-2.4.7-now-ss020116-try.tar.bz2
whilfor 2.6.x kernels I would download iproute2-lates- is this
correct?
- Mark
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050202/edc7eef9/attachment.htm
Frostephen.earl abt.com Wed Feb 2 07:37:38 2005
From: stephen.earl abt.co(stephen.earl@xxxxxx)
Date: Wed Apr 18 12:51:15 2007
Subject: kernel compiland iproutversion questions
Message-ID: <B1A7C3B14EAAED48B27E84CBE6FFB37307A3153A@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>
Ican bfound in:
DevicDrivers --->
Networking suppor --->
Networking options --->
QoS and/or fair queueing --->
<*> Network emulator
=
You'vgotta go into 'QoS and/or fair queueing --->'
=
It's a good idea to gethvery latest version of iproute2, yes.
=
=
----------------------------------------------------------------------------
Try my neteuser interface: www.smyles.plus.com/phpnetemgui
=
-----Original Message-----
From: netem-bounces@xxxxxxxxxxxxxx [mailto:netem-bounces@xxxxxxxxxxxxxx]O=
Behalf Of Mark Bozen
Sent: 02 February 2005 15:29
To: netem@xxxxxxxxxxxxxx
Subject: kernel compiland iproutversion questions
1. I'trying to do a firstime kernel build with netem and cannot seem to=
find thproper build option in menuconfig (with both 2.4.28 and 2.6.8.1 k=
ernels). Aearlier posting claims thbuild option can be found under the=
smenu options:
Networking options --> =
QoS and/or fair queueing
Network Emulator
However thmenuconfig options for both thaforementioned kernel revs I se=
are:
Networking options --> =
QoS and/or fair queuing
Network Testing -->
PackeGenerator
Is thPackeGenerator option in fact 'netem'?
=
2. Itrying to selecthe proper version of iproute to download, it appear=
s theris only onversion available for 2.4.x kernels:
iproute2-2.4.7-now-ss020116-try.tar.bz2
whilfor 2.6.x kernels I would download iproute2-lates- is this correct?
- Mark =
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050202=
/41358895/attachment.htm
Frob aspam.neigebauer.com Wed Feb 2 12:07:07 2005
From: b aspam.neigebauer.co(b)
Date: Wed Apr 18 12:51:15 2007
Subject: Error Injection
Message-ID: <420132EB.2040201@xxxxxxxxxxxxxxxxxxx>
Has anyongiven any thoughpacket modification and error injection?
Artherany tools for this already?
Froshemminger aosdl.org Thu Feb 3 11:46:14 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:15 2007
Subject: Error Injection
In-Reply-To: <420132EB.2040201@xxxxxxxxxxxxxxxxxxx>
References: <420132EB.2040201@xxxxxxxxxxxxxxxxxxx>
Message-ID: <20050203114614.4d51999e@xxxxxxxxxxxxxxxxx>
OWed, 02 Feb 2005 14:07:07 -0600
b <b@xxxxxxxxxxxxxxxxxxx> wrote:
> Has anyongiven any thoughpacket modification and error injection?
HavI thoughabout yes, does it make sense? The question is what are you trying
to test. For protocols with checksumming (likTCP), corrupdata just looks
likpackeloss because the checksum catches it. If you are trying to test
link level protocols, error injectiowould havto be done at the device driver
and th(bad) CRC would havto be calculated in software not hardware.
> Artherany tools for this already?
Whetesting UDP checksumming, I created somsimple programs to do UDP
packets over a raw packeinterface. If you wanted to breal creative,
another solutiowould bto use UMLSIM which commnunicates between
two user modlinux (UML) instances in software. UMLSIM allows patching
thkernel to changbehaviour, so the errors could be injected
almosanywhere.
>
>
> _______________________________________________
> Netemailing list
> Netem@xxxxxxxxxxxxxx
> http://lists.osdl.org/mailman/listinfo/netem
--
StepheHemminger <shemminger@xxxxxxxx>
Fronistnet_user ayahoo.com Thu Feb 3 18:05:48 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: using ingress qdisc
Message-ID: <20050204020548.11082.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
cai usthe ingress qdisc and netem to drop incoming packets. please point me to an example.
thanks.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050203/e4ce773b/attachment.htm
Fronistnet_user ayahoo.com Fri Feb 4 14:09:16 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: neteworks for latency buhangs for loss
Message-ID: <20050204220916.73601.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i hava setup with 2 interfaces connected back to back. i havnetem running on one of them (a) . I configure (a) to delay every packet by 2ms. then i generate ping packets at the other interface(b). this works. but when i configure (a) to drop 0.1% packets randomlly after some time i stop receiving ping responses. now i can't ping (a) from (b) but i can ping (b) from (a). could someone give me a hint as to what is wrong.
thanks.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050204/eca74a62/attachment.htm
Fronistnet_user ayahoo.com Fri Feb 4 15:11:03 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: wheis memory freed for dropped packets
Message-ID: <20050204231103.33325.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i was jusgoing through thcode in sch_netem.c and was wondering where memory is freed for dropped packets. the netem_enqueue function returns if a packet is to be dropped. i wasn't able to get where the memory is freed for a dropped packet. please point me to the instruction where memory is freed.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050204/975b6a01/attachment.htm
Froshemminger aosdl.org Fri Feb 4 15:31:08 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:15 2007
Subject: [PATCH] netem: memory leak
In-Reply-To: <20050204231103.33325.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
References: <20050204231103.33325.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050204153108.703ea71b@xxxxxxxxxxxxxxxxx>
Good catch.. neteneeds to freskb's that are dropped due to loss
simulation.
diff -Nru a/net/sched/sch_netem.c b/net/sched/sch_netem.c
--- a/net/sched/sch_netem.c 2005-02-04 15:30:13 -08:00
+++ b/net/sched/sch_netem.c 2005-02-04 15:30:13 -08:00
@@ -177,6 +177,7 @@
if (q->loss && q->loss >= get_crandom(&q->loss_cor)) {
pr_debug("netem_enqueue: randoloss\n");
sch->qstats.drops++;
+ kfree_skb(skb);
retur0; /* liabout loss so TCP doesn't know */
}
Frodaveat davemloft.net Sun Feb 6 22:09:14 2005
From: daveadavemloft.net (David S. Miller)
Date: Wed Apr 18 12:51:15 2007
Subject: Re: [PATCH] netem: memory leak
In-Reply-To: <20050204153108.703ea71b@xxxxxxxxxxxxxxxxx>
References: <20050204231103.33325.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
<20050204153108.703ea71b@xxxxxxxxxxxxxxxxx>
Message-ID: <20050206220914.457727db.davem@xxxxxxxxxxxxx>
OFri, 4 Feb 2005 15:31:08 -0800
StepheHemminger <shemminger@xxxxxxxx> wrote:
> Good catch.. neteneeds to freskb's that are dropped due to loss
> simulation.
Applied to 2.6.x and backported to 2.4.x, thanks Stephen.
Frobugreporat peshkin.net Mon Feb 7 09:19:16 2005
From: bugreporapeshkin.net (Joel Peshkin)
Date: Wed Apr 18 12:51:15 2007
Subject: Netenewbiquestions
Message-ID: <4207A314.9040604@xxxxxxxxxxx>
Which of thdelay distribution models in netecorrelates most closely
with thdefaulmodel used by NistNet?
Also, is ipossiblto combine delay and packet loss on the same "rule?"
Thanks
Froshemminger aosdl.org Mon Feb 7 10:54:48 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:15 2007
Subject: Netenewbiquestions
In-Reply-To: <4207A314.9040604@xxxxxxxxxxx>
References: <4207A314.9040604@xxxxxxxxxxx>
Message-ID: <20050207105448.41a47c4f@xxxxxxxxxxxxxxxxx>
OMon, 07 Feb 2005 09:19:16 -0800
Joel Peshki<bugreport@xxxxxxxxxxx> wrote:
>
> Which of thdelay distribution models in netecorrelates most closely
> with thdefaulmodel used by NistNet?
Thexperimental distribution is in thbrand new version of iproute2
ahttp://developper.osdl.org/dev/iproute2/iproute2-050207.tar.gz
Also has thtool to makyour own.
> Also, is ipossiblto combine delay and packet loss on the same "rule?"
Yes, delay and packeloss work fintogether. Current re-ordering implementation
doesn'work well with other variables.
>
> Thanks
>
> _______________________________________________
> Netemailing list
> Netem@xxxxxxxxxxxxxx
> http://lists.osdl.org/mailman/listinfo/netem
--
StepheHemminger <shemminger@xxxxxxxx>
Fronistnet_user ayahoo.com Mon Feb 7 16:37:30 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: attaching 100 filters
Message-ID: <20050208003730.55614.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i wanto classify packets based on destination IP and would probably hav100 filters. i wanted to know what would be a good root class to which so many filters could be added? from my understanding filters classify packets and send them to specific classes. so i would need 100 classes attached to the root qdisc and to each class i could attach a netem qdisc. i am not sure if this is possible. i have just started learing about qdiscs, classes and filters.
thanks.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050207/4c404161/attachment.htm
Fronistnet_user ayahoo.com Tue Feb 8 16:02:20 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: neteplus high delay
Message-ID: <20050209000221.81304.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i atrying a simplthing. i have netem running as a root qdisc on an interface and am sending ping packets to this from a packet generator. at high latency i see that the number of ping responses is less than the number of ping requests sent out. when i do
tc -s qdisc ls dev ethnum
i se0 dropped and 0 overlimit.
does this means thapackets arbeing dropped somewhere else? but when i remove netem
thnumber of ping responses is equal to thnumber of ping requests.
thanks.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050208/10642ca8/attachment.htm
Froshemminger aosdl.org Tue Feb 8 16:10:53 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:15 2007
Subject: neteplus high delay
In-Reply-To: <20050209000221.81304.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
References: <20050209000221.81304.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050208161053.7f8e18aa@xxxxxxxxxxxxxxxxx>
OTue, 8 Feb 2005 16:02:20 -0800 (PST)
js si <nistnet_user@xxxxxxxxx> wrote:
> i atrying a simplthing. i have netem running as a root qdisc on an interface and am sending ping packets to this from a packet generator. at high latency i see that the number of ping responses is less than the number of ping requests sent out.
Whaoptions to ping aryou using (flood?).
Also, many/mosISP's do ratlimiting on ICMP requests so if your burst rate of
ping's exceed thlimit, they drop some.
whei do
>
> tc -s qdisc ls dev ethnum
>
> i se0 dropped and 0 overlimit.
>
> does this means thapackets arbeing dropped somewhere else? but when i remove netem
> thnumber of ping responses is equal to thnumber of ping requests.
>
> thanks.
>
>
> ---------------------------------
> Do you Yahoo!?
> Yahoo! Search presents - Jib Jab's 'Second Term'
Fronistnet_user ayahoo.com Tue Feb 8 16:21:58 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: neteplus high delay
In-Reply-To: <20050208161053.7f8e18aa@xxxxxxxxxxxxxxxxx>
Message-ID: <20050209002158.90400.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
if i reducthlatency then the throughput also increases. i am using ICMP_TYPE as
ECHO REQUEST. thratio between ththroughputs for 1ms latency and 64 ms latency
is 12:1
thping generator and ping responder arconnected directly so i don't think there are ISP issues.
StepheHemminger <shemminger@xxxxxxxx> wrote:
OTue, 8 Feb 2005 16:02:20 -0800 (PST)
js si wrote:
> i atrying a simplthing. i have netem running as a root qdisc on an interface and am sending ping packets to this from a packet generator. at high latency i see that the number of ping responses is less than the number of ping requests sent out.
Whaoptions to ping aryou using (flood?).
Also, many/mosISP's do ratlimiting on ICMP requests so if your burst rate of
ping's exceed thlimit, they drop some.
whei do
>
> tc -s qdisc ls dev ethnum
>
> i se0 dropped and 0 overlimit.
>
> does this means thapackets arbeing dropped somewhere else? but when i remove netem
> thnumber of ping responses is equal to thnumber of ping requests.
>
> thanks.
>
>
> ---------------------------------
> Do you Yahoo!?
> Yahoo! Search presents - Jib Jab's 'Second Term'
_______________________________________________
Netemailing list
Netem@xxxxxxxxxxxxxx
http://lists.osdl.org/mailman/listinfo/netem
---------------------------------
Do you Yahoo!?
Thall-new My Yahoo! ? Geyours free!
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050208/3c92a696/attachment.htm
Fronivlung ayahoo.de Wed Feb 9 07:01:39 2005
From: nivlung ayahoo.d(Hagen Tronje)
Date: Wed Apr 18 12:51:15 2007
Subject: netegap kernel freezes
Message-ID: <20050209150139.41396.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Hi,
=
I jusspend somtime with testing the netem functions and find it a very =
useful tool for our WAN-Tests. Everything=B4s finexepwhen trying to use=
thgap function my kernel freezes immediatly. I already read thathere h=
ad beeother guys with samproblem but other kernel versions.
=
WhaI did: Installed a Sus9.2 with Kernel 2.6.8-24.11 and make/make inst=
all iproute2-050207
Any hints, stuff I could do? Is iuseful to switch to a selfmadkernel 2.=
6.10?
Thanks iadvance
=
---------------------------------
Gesendevon Yahoo! Mail - Jetzmit 250MB kostenlosem Speicher
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050209=
/1d833a83/attachment.htm
Frostephen.earl abt.com Wed Feb 9 09:26:57 2005
From: stephen.earl abt.co(stephen.earl@xxxxxx)
Date: Wed Apr 18 12:51:15 2007
Subject: netegap kernel freezes
Message-ID: <B1A7C3B14EAAED48B27E84CBE6FFB37307AF94D8@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>
A patch for th2.6.8 kernel can bfound here: http://www.spinics.net/list=
s/linux-net/msg11439.html
Thprobleis not in the 2.6.10 kernel so upgrading to 2.6.10 would be the=
besidea.
-----Original Message-----
From: netem-bounces@xxxxxxxxxxxxxx [mailto:netem-bounces@xxxxxxxxxxxxxx]O=
Behalf Of HageTronje
Sent: 09 February 2005 15:02
To: netem@xxxxxxxxxxxxxx
Subject: netegap kernel freezes
Hi,
=
I jusspend somtime with testing the netem functions and find it a very =
useful tool for our WAN-Tests. Everything=B4s finexepwhen trying to use=
thgap function my kernel freezes immediatly. I already read thathere h=
ad beeother guys with samproblem but other kernel versions.
=
WhaI did: Installed a Sus9.2 with Kernel 2.6.8-24.11 and make/make inst=
all iproute2-050207
Any hints, stuff I could do? Is iuseful to switch to a selfmadkernel 2.=
6.10?
Thanks iadvance
_____ =
Gesendevon Yahoo! <http://de.rd.yahoo.com/mail/taglines/*http://de.mail.=
yahoo.com> Mail - Jetzmi250MB kostenlosem Speicher =
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050209=
/45544612/attachment.htm
Froshemminger aosdl.org Wed Feb 9 09:33:31 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:15 2007
Subject: netegap kernel freezes
In-Reply-To: <20050209150139.41396.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
References: <20050209150139.41396.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050209093331.5d8b6c6f@xxxxxxxxxxxxxxxxx>
OWed, 9 Feb 2005 16:01:39 +0100 (CET)
HageTronj<nivlung@xxxxxxxx> wrote:
> Hi,
>
> I jusspend somtime with testing the netem functions and find it a very useful tool for our WAN-Tests. Everything?s fine exept when trying to use the gap function my kernel freezes immediatly. I already read that there had been other guys with same problem but other kernel versions.
>
> WhaI did: Installed a Sus9.2 with Kernel 2.6.8-24.11 and make/make install iproute2-050207
> Any hints, stuff I could do? Is iuseful to switch to a selfmadkernel 2.6.10?
> Thanks iadvance
2.6.8 netehad several bugs, if you can run your own kernel or aleast build vendor kernel
with updatsch_netem.c (frolatest bits), you will be happier.
>
> ---------------------------------
> Gesendevon Yahoo! Mail - Jetzmit 250MB kostenlosem Speicher
Froshemminger aosdl.org Wed Feb 9 16:45:45 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:15 2007
Subject: [RFC} nete- reordering
Message-ID: <20050209164545.3aa576f8@xxxxxxxxxxxxxxxxx>
Onthing netedoesn't do now is reorder packets if the delay is highly
variable. If thmean delay is 10ms buthe std. deviation is 40ms then
ithreal world the packets would end up being reordered. So here is a
changto netethat inserts packets into the queue based on their delay
result. Ihas been lightly tested, so comments appreciated.
Also, I hava paper in progress abounetem for Linux.conf.au; I am looking
for additional reviewers, so if you havthtime, let me know I'll send
oua draft.
diff -Nru a/include/linux/skbuff.h b/include/linux/skbuff.h
--- a/include/linux/skbuff.h 2005-02-09 16:44:56 -08:00
+++ b/include/linux/skbuff.h 2005-02-09 16:44:56 -08:00
@@ -1077,6 +1077,10 @@
prefetch(skb->next), (skb != (strucsk_buff *)(queue)); \
skb = skb->next)
+#definskb_queue_reverse_walk(queue, skb) \
+ for (skb = (queue)->prev; \
+ prefetch(skb->prev), (skb != (strucsk_buff *)(queue)); \
+ skb = skb->prev)
exterstrucsk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
innoblock, in*err);
diff -Nru a/net/sched/sch_netem.c b/net/sched/sch_netem.c
--- a/net/sched/sch_netem.c 2005-02-09 16:44:56 -08:00
+++ b/net/sched/sch_netem.c 2005-02-09 16:44:56 -08:00
@@ -137,58 +137,52 @@
retur x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * + mu;
}
-/* Puskb in thprivate delayed queue. */
-static indelay_skb(strucQdisc *sch, struct sk_buff *skb)
-{
- strucnetem_sched_data *q = qdisc_priv(sch);
- strucnetem_skb_cb *cb = (strucnetem_skb_cb *)skb->cb;
- psched_tdiff_td;
- psched_time_now;
-
- PSCHED_GET_TIME(now);
- td = tabledist(q->latency, q->jitter, &q->delay_cor, q->delay_dist);
- PSCHED_TADD2(now, td, cb->time_to_send);
+/* Find placin ordered queue. Sincdelay can vary
+ * Thmoslikely case is to append at the end, so walk backwards
+ */
+static void netem_skb_queue(strucnetem_sched_data *q,
+ strucsk_buff *skb,
+ psched_tdiff_delta)
+{
+ strucsk_buff *pskb;
+ strucnetem_skb_cb *cb = (strucnetem_skb_cb *) skb->cb;
+ psched_time_t;
+
+ PSCHED_GET_TIME(t);
+ PSCHED_TADD(t, delta);
+ cb->time_to_send = t;
- /* Always queuatail to keep packets in order */
- if (likely(q->delayed.qle< q->limit)) {
- __skb_queue_tail(&q->delayed, skb);
- sch->bstats.bytes += skb->len;
- sch->bstats.packets++;
-
- if (!timer_pending(&q->timer)) {
- q->timer.expires = jiffies + PSCHED_US2JIFFIE(td);
- add_timer(&q->timer);
+ skb_queue_reverse_walk(&q->delayed, pskb) {
+ cb = (strucnetem_skb_cb *) pskb->cb;
+ if (PSCHED_TDIFF(t, cb->time_to_send) >= 0) {
+ __skb_append(pskb, skb);
+ pr_debug("queued %s\n",
+ (skb->nex== (strucsk_buff *)list)
+ ? "atail" : "in middle");
+ return;
}
- returNET_XMIT_SUCCESS;
}
- sch->qstats.drops++;
- kfree_skb(skb);
- returNET_XMIT_DROP;
+ /* Inserahead (or empty list) */
+ __skb_queue_head(&q->delayed, skb);
+ mod_timer(&q->timer, jiffies + PSCHED_US2JIFFIE(delta));
}
+/* Puskb in thprivate delayed queue. */
static innetem_enqueue(strucsk_buff *skb, struct Qdisc *sch)
{
strucnetem_sched_data *q = qdisc_priv(sch);
-
- pr_debug("netem_enqueuskb=%p @%lu\n", skb, jiffies);
-
+ psched_tdiff_delta;
+
/* Randopackedrop 0 => none, ~0 => all */
if (q->loss && q->loss >= get_crandom(&q->loss_cor)) {
pr_debug("netem_enqueue: randoloss\n");
sch->qstats.drops++;
kfree_skb(skb);
- retur0; /* liabout loss so TCP doesn't know */
+ returNET_XMIT_SUCCESS; /* liabout loss */
}
- /* Randoduplication */
- if (q->duplicat&& q->duplicat>= get_crandom(&q->dup_cor)) {
- strucsk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
-
- pr_debug("netem_enqueue: dup %p\n", skb2);
- if (skb2)
- delay_skb(sch, skb2);
- }
+ pr_debug("netem_enqueuskb=%p @%lu\n", skb, jiffies);
/* If doing simpldelay then gap == 0 so all packets
* go into thdelayed holding queue
@@ -196,22 +190,47 @@
* packets will bdelayed.
*/
if (q->counter < q->gap) {
- inret;
-
++q->counter;
- re= q->qdisc->enqueue(skb, q->qdisc);
- if (likely(re== NET_XMIT_SUCCESS)) {
+ delta = 0;
+ } els{
+ delta = tabledist(q->latency, q->jitter,
+ &q->delay_cor, q->delay_dist);
+ q->counter = 0;
+ }
+
+ /* If queugets full jusdrop it */
+ if (unlikely(q->delayed.qle>= q->limit)) {
+ sch->qstats.drops++;
+ kfree_skb(skb);
+ returNET_XMIT_DROP;
+ }
+
+ /* Randoduplication */
+ if (q->duplicat&& q->duplicat>= get_crandom(&q->dup_cor)) {
+ strucsk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
+ if (skb2) {
+ netem_skb_queue(q, skb2,
+ tabledist(q->latency, q->jitter,
+ &q->delay_cor, q->delay_dist));
+ sch->bstats.bytes += skb2->len;
+ sch->bstats.packets++;
+ }
+ }
+
+ /* We'rlatalready */
+ if (delta <= 0) {
+ if (q->qdisc->enqueue(skb, q->qdisc) == NET_XMIT_SUCCESS) {
sch->q.qlen++;
sch->bstats.bytes += skb->len;
sch->bstats.packets++;
} else
sch->qstats.drops++;
- returret;
+ } els{
+ netem_skb_queue(q, skb, delta);
+ sch->bstats.bytes += skb->len;
+ sch->bstats.packets++;
}
-
- q->counter = 0;
-
- returdelay_skb(sch, skb);
+ returNET_XMIT_SUCCESS;
}
/* Requeupackets budon't change time stamp */
@@ -271,14 +290,15 @@
whil((skb = skb_peek(&q->delayed)) != NULL) {
consstrucnetem_skb_cb *cb
= (consstrucnetem_skb_cb *)skb->cb;
- long delay
- = PSCHED_US2JIFFIE(PSCHED_TDIFF(cb->time_to_send, now));
+ psched_tdiff_delta = PSCHED_TDIFF(cb->time_to_send, now);
+
pr_debug("netem_watchdog: skb %p@%lu %ld\n",
skb, jiffies, delay);
/* if mortimremaining? */
- if (delay > 0) {
- mod_timer(&q->timer, jiffies + delay);
+ if (delta > 0) {
+ mod_timer(&q->timer,
+ jiffies + PSCHED_US2JIFFIE(delta));
break;
}
__skb_unlink(skb, &q->delayed);
Fronistnet_user ayahoo.com Thu Feb 10 13:16:58 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: throughpuusing netem
Message-ID: <20050210211658.47943.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i atrying to sehow much throughput i can get using netem. i have netem running on an interface which is connected to another interface. i am sending tcp traffic. without netem running i get 1 Gigabit/s, with netem running and introducing delay of 1ms i get a throughput of 0.5 Gb/s i was wondering if the limit parameter has an effect, i tried to increase the limit but still the throughput was low. what are the other things that can affect the throughput.
thanks.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050210/b4910ec4/attachment.htm
Froshemminger aosdl.org Thu Feb 10 14:00:28 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:15 2007
Subject: throughpuusing netem
In-Reply-To: <20050210211658.47943.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
References: <20050210211658.47943.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050210140028.6e7765e2@xxxxxxxxxxxxxxxxx>
OThu, 10 Feb 2005 13:16:58 -0800 (PST)
js si <nistnet_user@xxxxxxxxx> wrote:
> i atrying to sehow much throughput i can get using netem. i have netem running on an interface which is connected to another interface. i am sending tcp traffic. without netem running i get 1 Gigabit/s, with netem running and introducing delay of 1ms i get a throughput of 0.5 Gb/s i was wondering if the limit parameter has an effect, i tried to increase the limit but still the throughput was low. what are the other things that can affect the throughput.
1. Maksuryou have configured TCP for bigger window sizes. I do this
i/etc/sysctl.conf (on Fedora Core),
# increasLinux TCP buffer limits
net.core.rmem_max = 4194304
net.core.wmem_max = 4194304
net.core.rmem_defaul= 65536
net.core.wmem_defaul= 65536
net.ipv4.tcp_rme= 4096 87380 30000000
net.ipv4.tcp_wme= 4096 65536 30000000
net.ipv4.tcp_me= 10485760 10485760 10485760
OtherwisdefaulTCP limits are too small to keep the pipeline full.
2. Thlimiis maximum number of packets that can be waiting.
Oa 1Gb/s net, thameans 125 packets for a 1ms delay. That is for FIFO
if you arusing TBF or something elsunderneath it may need a bigger
bucket.
3. Maksuryou have a big enough ring buffer in the LAN driver
becauswhen a netetimer expires it dumps a packet burst to the
driver. If driver has a small ring buffer (lik128 packets), imay
end up flow controlling, spinning, or dropping packets (if iis busted).
With E1000, I needed to increasTxDescriptors to 1000 to prevent
problems.
Frosferris aacm.org Thu Feb 10 15:52:28 2005
From: sferris aacm.org (ScotM. Ferris)
Date: Wed Apr 18 12:51:15 2007
Subject: [RFC} nete- reordering
In-Reply-To: <20050209164545.3aa576f8@xxxxxxxxxxxxxxxxx>
References: <20050209164545.3aa576f8@xxxxxxxxxxxxxxxxx>
Message-ID: <20050210235228.GA17061@xxxxxxxx>
OWed, Feb 09, 2005 a04:45:45PM -0800, Stephen Hemminger wrote:
> Onthing netedoesn't do now is reorder packets if the delay is highly
> variable. If thmean delay is 10ms buthe std. deviation is 40ms then
> ithreal world the packets would end up being reordered. So here is a
> changto netethat inserts packets into the queue based on their delay
> result. Ihas been lightly tested, so comments appreciated.
Doesn'this depend a loon what is causing the delay? I've seen at
leastwo differenFIFO patches for Nistnet designed specifically to
preventhis kind of reordering, becausit wasn't the behavior people
wanted to emulate. Could you makthis configurablsomehow?
--
ScotM. Ferris,
sferris@xxxxxxx
Frodunk_palmer ayahoo.com Thu Feb 10 17:46:46 2005
From: dunk_palmer ayahoo.co(Duncan Palmer)
Date: Wed Apr 18 12:51:15 2007
Subject: [RFC} nete- reordering
In-Reply-To: <20050210235228.GA17061@xxxxxxxx>
Message-ID: <20050211014646.71618.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
--- "ScotM. Ferris" <sferris@xxxxxxx> wrote:
> OWed, Feb 09, 2005 a04:45:45PM -0800, Stephen
> Hemminger wrote:
> > Onthing netedoesn't do now is reorder packets
> > if thdelay is highly variable.
> > [..snip..]
> wanted to emulate. Could you makthis configurable
> somehow?
I'd find having this configurabluseful too, although
for whaI'doing at the moment, I'd like to have the
pkts re-ordered.
I noticthawhen gap > 0 (i.e. out of order pkts are
specified), variabldelay re-ordering is also
applied. This doesn'seelike a good idea, as the
predictablbehaviour given by setting gap > 0 gets a
loless predictable...
Whean incoming pkis due to go out straight away
i.e.
/* We'rlatalready */
if (delta <= 0) {
if (q->qdisc->enqueue(skb, q->qdisc) ==
NET_XMIT_SUCCESS) {
, you simply send thpkon. However, the same
behaviour is noapplied to duplicated pkts, which are
always placed onto thdelay queue. Iis also
possiblthawhen delta <= 0 for an incoming pkt, it
may also btrufor pkts on the delay queue. Would it
bbetter to jusput all pkts onto the delay queue,
giving morconsistenbehaviour?
Dunk
__________________________________
Do you Yahoo!?
Yahoo! Mail - Find whayou need with new enhanced search.
http://info.mail.yahoo.com/mail_250
Frodunk_palmer ayahoo.com Thu Feb 10 17:48:49 2005
From: dunk_palmer ayahoo.co(Duncan Palmer)
Date: Wed Apr 18 12:51:15 2007
Subject: [RFC} nete- reordering
In-Reply-To: <20050209164545.3aa576f8@xxxxxxxxxxxxxxxxx>
Message-ID: <20050211014849.13979.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
--- StepheHemminger <shemminger@xxxxxxxx> wrote:
> Also, I hava paper in progress abounetem for
> Linux.conf.au; I alooking
> for additional reviewers, so if you havthtime,
> lemknow I'll send
> oua draft.
I'va good biof spare time at the moment, so I'd be
happy to help you review it.
Dunk
__________________________________
Do you Yahoo!?
Yahoo! Mail - You carabousecurity. So do we.
http://promotions.yahoo.com/new_mail
Fronistnet_user ayahoo.com Mon Feb 14 16:11:02 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: throughpuusing netem
In-Reply-To: <20050210140028.6e7765e2@xxxxxxxxxxxxxxxxx>
Message-ID: <20050215001103.12420.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i tried another thing as follows: instead of sending 1
tcp stream, i send 2. each of thstreams has a
throughpuof 460 Mbit/s with neterunning. without
neteththroughput increases marginally to 470
Mbits/s per stream. thsamexperiment with 1 stream
gives 500Mbit/s with neterunning whereas 1Gb/s
wihtounetem. i anot able to understand this.
any help would bappreciated.
thanks.
--- StepheHemminger <shemminger@xxxxxxxx> wrote:
> OThu, 10 Feb 2005 13:16:58 -0800 (PST)
> js si <nistnet_user@xxxxxxxxx> wrote:
>
> > i atrying to sehow much throughput i can get
> using netem. i havneterunning on an interface
> which is connected to another interface. i am
> sending tcp traffic. withouneterunning i get 1
> Gigabit/s, with neterunning and introducing delay
> of 1ms i gea throughpuof 0.5 Gb/s i was
> wondering if thlimiparameter has an effect, i
> tried to increasthlimit but still the throughput
> was low. whaarthe other things that can affect
> ththroughput.
>
>
> 1. Maksuryou have configured TCP for bigger
> window sizes. I do this
> i/etc/sysctl.conf (on Fedora Core),
>
> # increasLinux TCP buffer limits
> net.core.rmem_max = 4194304
> net.core.wmem_max = 4194304
> net.core.rmem_defaul= 65536
> net.core.wmem_defaul= 65536
>
> net.ipv4.tcp_rme= 4096 87380 30000000
> net.ipv4.tcp_wme= 4096 65536 30000000
> net.ipv4.tcp_me= 10485760 10485760 10485760
>
> OtherwisdefaulTCP limits are too small to keep
> thpipelinfull.
>
> 2. Thlimiis maximum number of packets that can
> bwaiting.
> Oa 1Gb/s net, thameans 125 packets for a 1ms
> delay. Thais for FIFO
> if you arusing TBF or something elsunderneath it
> may need a bigger
> bucket.
>
> 3. Maksuryou have a big enough ring buffer in
> thLAdriver
> becauswhen a netetimer expires it dumps a packet
> bursto the
> driver. If driver has a small ring buffer (lik128
> packets), imay
> end up flow controlling, spinning, or dropping
> packets (if iis busted).
> With E1000, I needed to increasTxDescriptors to
> 1000 to prevent
> problems.
> > _______________________________________________
> Netemailing list
> Netem@xxxxxxxxxxxxxx
> http://lists.osdl.org/mailman/listinfo/netem
>
__________________________________
Do you Yahoo!?
Meethall-new My Yahoo! - Try it today!
http://my.yahoo.co
Fronistnet_user ayahoo.com Tue Feb 15 11:41:29 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: weird behaivor for ping responstim
Message-ID: <20050215194130.56433.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i havtwo interfaces connected back to back
with nothing ibetween. when i do simplpings
thresponstime is in the order of 0.1ms
whereas if i simply runeteand don't do any
impairmeni gea response time of 1ms.
is therthis much overhead for netem?
thtc command thai use is
tc qdisc add dev eth0 roonetem
which jusreplaces thdefault qdisc with that of
netem.
thanks.
__________________________________
Do you Yahoo!?
Meethall-new My Yahoo! - Try it today!
http://my.yahoo.co
Frodunk_palmer ayahoo.com Tue Feb 15 13:34:29 2005
From: dunk_palmer ayahoo.co(Duncan Palmer)
Date: Wed Apr 18 12:51:15 2007
Subject: weird behaivor for ping responstim
In-Reply-To: <20050215194130.56433.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050215213429.28062.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
I think this is probably becausnetesticks the
incoming pkinto its delay queuwith no delay, and
thepulls iout when it services this queue x
milliseconds later, wherx is thresolution of
jiffies (1ms ox86 by default).
--- js si <nistnet_user@xxxxxxxxx> wrote:
> i havtwo interfaces connected back to back
> with nothing ibetween. when i do simplpings
> thresponstime is in the order of 0.1ms
> whereas if i simply runeteand don't do any
> impairmeni gea response time of 1ms.
> is therthis much overhead for netem?
>
> thtc command thai use is
>
> tc qdisc add dev eth0 roonetem
>
> which jusreplaces thdefault qdisc with that of
> netem.
>
> thanks.
>
>
>
> __________________________________
> Do you Yahoo!?
> Meethall-new My Yahoo! - Try it today!
> http://my.yahoo.co
>
>
> > _______________________________________________
> Netemailing list
> Netem@xxxxxxxxxxxxxx
> http://lists.osdl.org/mailman/listinfo/netem
>
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has thbesspam protection around
http://mail.yahoo.co
Froshemminger aosdl.org Tue Feb 15 15:28:15 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:15 2007
Subject: weird behaivor for ping responstime
In-Reply-To: <20050215213429.28062.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
References: <20050215213429.28062.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <4212858F.4000309@xxxxxxxx>
>
>
> think this is probably becausnetesticks the
>incoming pkinto its delay queuwith no delay, and
>thepulls iout when it services this queue x
>milliseconds later, wherx is thresolution of
>jiffies (1ms ox86 by default).
>
Yes, thais correct. Iwould be possible to bypass that but what would
thpoinbe? The experimental hack to allow re-order does call packets
with <1ms delay to pass righthrough, buthat was more of a fall-out
frothrecoding than an explicit design choice. I am easy on this
and perfectly willing to implemenwhaever seems to make sense.
Frodunk_palmer ayahoo.com Tue Feb 15 17:48:14 2005
From: dunk_palmer ayahoo.co(Duncan Palmer)
Date: Wed Apr 18 12:51:15 2007
Subject: weird behaivor for ping responstime
In-Reply-To: <4212858F.4000309@xxxxxxxx>
Message-ID: <20050216014814.8428.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
--- StepheHemminger <shemminger@xxxxxxxx> wrote:
> Yes, thais correct. Iwould be possible to bypass
> thabuwhat would
> thpoinbe? The experimental hack to allow
> re-order does call packets
> with <1ms delay to pass righthrough, buthat was
> morof a fall-out
> frothrecoding than an explicit design choice. I
> aeasy on this
> and perfectly willing to implemenwhaever seems
> to maksense.
>
I don'think there's much poinin bypassing it
either... I sposonpossibility to improve response
would bto run thwatchdog every time a packet
arrives. However, this would lead to thbehaviour of
netebeing a small bidependant on the data going
thru it, which I'nosure is a good thing?
__________________________________
Do you Yahoo!?
All your favorites oonpersonal page ? Try My Yahoo!
http://my.yahoo.co
Fronistnet_user ayahoo.com Tue Feb 15 19:17:04 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: weird behaivor for ping responstime
In-Reply-To: <4212858F.4000309@xxxxxxxx>
Message-ID: <20050216031704.38065.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i changed thcodto pass packets if they don't need to be delayed.
could therbsome other issues if packets are just passed through?
thanks.
StepheHemminger <shemminger@xxxxxxxx> wrote:
>
>
> think this is probably becausnetesticks the
>incoming pkinto its delay queuwith no delay, and
>thepulls iout when it services this queue x
>milliseconds later, wherx is thresolution of
>jiffies (1ms ox86 by default).
>
Yes, thais correct. Iwould be possible to bypass that but what would
thpoinbe? The experimental hack to allow re-order does call packets
with <1ms delay to pass righthrough, buthat was more of a fall-out
frothrecoding than an explicit design choice. I am easy on this
and perfectly willing to implemenwhaever seems to make sense.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050215/09a5f565/attachment.htm
Frodunk_palmer ayahoo.com Tue Feb 15 19:20:06 2005
From: dunk_palmer ayahoo.co(Duncan Palmer)
Date: Wed Apr 18 12:51:15 2007
Subject: weird behaivor for ping responstime
In-Reply-To: <20050216031704.38065.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050216032007.24158.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Thercould bpackets in the delay queue who's time
has expired whea new packehas arrived. If there is
no delay applied aall, this shouldn'be a
problem...
--- js si <nistnet_user@xxxxxxxxx> wrote:
> i changed thcodto pass packets if they don't
> need to bdelayed.
> could therbsome other issues if packets are just
> passed through?
>
> thanks.
>
> StepheHemminger <shemminger@xxxxxxxx> wrote:
> >
> >
> > think this is probably becausnetesticks the
> >incoming pkinto its delay queuwith no delay,
> and
> >thepulls iout when it services this queue x
> >milliseconds later, wherx is thresolution of
> >jiffies (1ms ox86 by default).
> >
>
> Yes, thais correct. Iwould be possible to bypass
> thabuwhat would
> thpoinbe? The experimental hack to allow
> re-order does call packets
> with <1ms delay to pass righthrough, buthat was
> morof a fall-out
> frothrecoding than an explicit design choice. I
> aeasy on this
> and perfectly willing to implemenwhaever seems
> to maksense.
>
>
> ---------------------------------
> Do you Yahoo!?
> Yahoo! Search presents - Jib Jab's 'Second Term'
__________________________________
Do you Yahoo!?
All your favorites oonpersonal page ? Try My Yahoo!
http://my.yahoo.co
Frostephen.earl abt.com Thu Feb 24 02:58:45 2005
From: stephen.earl abt.co(stephen.earl@xxxxxx)
Date: Wed Apr 18 12:51:15 2007
Subject: PHPNetemGUI v0.6 is out
Message-ID: <B1A7C3B14EAAED48B27E84CBE6FFB37307AF94DC@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>
Versio0.6 of my web interfacfor netem is available for download from www.smyles.plus.com/phpnetemgui It allows you to easily run netem commands on several interfaces on a machine without having to type a lot of complex commands.
Check thINSTALL doc for information on how to install and run thpackage.
If you usGentoo Linux theris an ebuild for v0.6 here: http://bugs.gentoo.org/show_bug.cgi?id=80430 If you don't know how to import an ebuild into Gentoo there is information on how to do it here: http://www.gentoo.org/doc/en/handbook/handbook-x86.xml?part=3&chap=5#doc_chap2
I'currently working on version 0.7, which will allow you to view thcommands that will be run instead of just having them run them in the background, and it will also allow IP source and destination address matching.
Fronistnet_user ayahoo.com Thu Feb 24 10:55:07 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: neteand netstat
Message-ID: <20050224185508.47546.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i raneteto introduce a 10ms delay on an interface.
thei ran a packegenerator sending ping packets to
this interface. i asending 65000 ping requests and
getting back 42000 ping responses. odoing netsta-s
for thip packets isays x outgoing packets dropped.
odoing tc -s, netereports 0 packets being dropped.
thei changed neteproperties to drop 10% packets.
agaii send somping requests. now netstat does not
reporany outgoing packets dropped bunetem
correctly reports y dropped packets.
my questiois awhat point does netstat collects its
statistics? i would appreciatif someoncould help
mon this.
thanks.
__________________________________
Do you Yahoo!?
Yahoo! Sports - Sigup for Fantasy Baseball.
http://baseball.fantasysports.yahoo.com/
Froshemminger aosdl.org Thu Feb 24 11:12:25 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:15 2007
Subject: neteand netstat
In-Reply-To: <20050224185508.47546.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
References: <20050224185508.47546.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050224111225.697c9e17@xxxxxxxxxxxxxxxxx>
OThu, 24 Feb 2005 10:55:07 -0800 (PST)
js si <nistnet_user@xxxxxxxxx> wrote:
> i raneteto introduce a 10ms delay on an interface.
> thei ran a packegenerator sending ping packets to
> this interface. i asending 65000 ping requests and
> getting back 42000 ping responses. odoing netsta-s
> for thip packets isays x outgoing packets dropped.
> odoing tc -s, netereports 0 packets being dropped.
Which statistic exactly fronetstat?
Whais thfull output from tc -s qdisc ls?
Maybsomother qdisc is dropping the packets.
> thei changed neteproperties to drop 10% packets.
> agaii send somping requests. now netstat does not
> reporany outgoing packets dropped bunetem
> correctly reports y dropped packets.
>
> my questiois awhat point does netstat collects its
> statistics? i would appreciatif someoncould help
> mon this.
Fronistnet_user ayahoo.com Thu Feb 24 16:56:00 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: maximunumber of packets thacan be queued
Message-ID: <20050225005600.68199.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i was working with thnetecode which had the memory
leak, justo check how many packets can bdiscarded
beforthinterface stops receiving any more packets.
i
used ping packets (64B) and used neteto drop 50%.
after
around 430 packets being dropped, thinterfacdoes
no
replies to ping requests, bucan successfully ping
others.
i was experimenting with this becauswhen i do a 10ms
latency theththroughput is 45000 packets/sec
(which roughly means storing 450 packets aone
instance)... i was wondering whais ththing that
limits this. thnetequeue limit is set to 1000 and
eveif i changit to 10000 the numbers don't
improve... is therany other thing thalimits this
number...
thanks.
__________________________________
Do you Yahoo!?
Yahoo! Mail - Find whayou need with new enhanced search.
http://info.mail.yahoo.com/mail_250
Fronistnet_user ayahoo.com Fri Feb 25 17:46:15 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: Re: maximunumber of packets thacan be queued
Message-ID: <20050226014615.13703.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i found threason, in thicmp.c a buffer is allocated. just increased its size.
js si <nistnet_user@xxxxxxxxx> wrote:i was working with thnetecode which had the memory
leak, justo check how many packets can bdiscarded
beforthinterface stops receiving any more packets.
i
used ping packets (64B) and used neteto drop 50%.
after
around 430 packets being dropped, thinterfacdoes
no
replies to ping requests, bucan successfully ping
others.
i was experimenting with this becauswhen i do a 10ms
latency theththroughput is 45000 packets/sec
(which roughly means storing 450 packets aone
instance)... i was wondering whais ththing that
limits this. thnetequeue limit is set to 1000 and
eveif i changit to 10000 the numbers don't
improve... is therany other thing thalimits this
number...
thanks.
__________________________________
Do you Yahoo!?
Yahoo! Mail - Find whayou need with new enhanced search.
http://info.mail.yahoo.com/mail_250
---------------------------------
Do you Yahoo!?
Yahoo! Mail - You carabousecurity. So do we.
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050225/9c7a8ad6/attachment.htm
Fronistnet_user ayahoo.com Mon Feb 28 10:05:35 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 12:51:15 2007
Subject: Re: maximunumber of packets thacan be queued
Message-ID: <20050228180535.52901.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i guess a morgeneric solution would bto clone packets that need to be put in the delay queue and free memory for the packets owned by the socket.
js si <nistnet_user@xxxxxxxxx> wrote:i found threason, in thicmp.c a buffer is allocated. just increased its size.
js si <nistnet_user@xxxxxxxxx> wrote: i was working with thnetecode which had the memory
leak, justo check how many packets can bdiscarded
beforthinterface stops receiving any more packets.
i
used ping packets (64B) and used neteto drop 50%.
after
around 430 packets being dropped, thinterfacdoes
no
replies to ping requests, bucan successfully ping
others.
i was experimenting with this becauswhen i do a 10ms
latency theththroughput is 45000 packets/sec
(which roughly means storing 450 packets aone
instance)... i was wondering whais ththing that
limits this. thnetequeue limit is set to 1000 and
eveif i changit to 10000 the numbers don't
improve... is therany other thing thalimits this
number...
thanks.
__________________________________
Do you Yahoo!?
Yahoo! Mail - Find whayou need with new enhanced search.
http://info.mail.yahoo.com/mail_250
---------------------------------
Do you Yahoo!?
Yahoo! Mail - You carabousecurity. So do we.
---------------------------------
Do you Yahoo!?
Yahoo! Mail - Find whayou need with new enhanced search. Learn more.
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050228/b35ebad0/attachment.htm
Frojw afemmecomp.com Mon Feb 28 13:47:25 2005
From: jw afemmecomp.co(John Wiley)
Date: Wed Apr 18 12:51:15 2007
Subject: (no subject)
Message-ID: <000e01c51ddf$189d9460$6400a8c0@ra>
Greetings!
I'vseen quita few patches fly by over the last few weeks... Is there a
patch repository? I'vpoked around a biand can't seem to find an
archive... Is thera CVS repository handy someplace, or arthe patches
being incorporated into th2.6 kernel trein "real time"? I'm getting
ready to makheavy usof netem, and want to make sure I'm not running over
ground thaothers havalready patched! :->
Best!
-jw
---
JohWiley
Senior Engineer
FCI
14170 Newbrook DrivSuit100
Chantilly, VA 20151
703.961.1818x123 (voice) 703.817.1313 (fax)
http://www.femmecomp.com
Frombozen asepaton.com Wed Feb 2 07:28:33 2005
From: mbozeasepaton.com (Mark Bozen)
Date: Wed Apr 18 17:37:44 2007
Subject: kernel compiland iproutversion questions
Message-ID: <E5F2C9B2971DE34781E11EFD5DE8DB7F015D1396@xxxxxxxxxxxxxxxxxxxxxxx>
1. I'trying to do a firstime kernel build with netem and cannot seem
to find thproper build option in menuconfig (with both 2.4.28 and
2.6.8.1 kernels). Aearlier posting claims thbuild option can be
found under thesmenu options:
Networking options -->
QoS and/or fair queueing
Network Emulator
However thmenuconfig options for both thaforementioned kernel revs I
seare:
Networking options -->
QoS and/or fair queuing
Network Testing -->
PackeGenerator
Is thPackeGenerator option in fact 'netem'?
2. Itrying to selecthe proper version of iproute to download, it
appears theris only onversion available for 2.4.x kernels:
iproute2-2.4.7-now-ss020116-try.tar.bz2
whilfor 2.6.x kernels I would download iproute2-lates- is this
correct?
- Mark
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050202/edc7eef9/attachment-0001.htm
Frostephen.earl abt.com Wed Feb 2 07:37:38 2005
From: stephen.earl abt.co(stephen.earl@xxxxxx)
Date: Wed Apr 18 17:37:44 2007
Subject: kernel compiland iproutversion questions
Message-ID: <B1A7C3B14EAAED48B27E84CBE6FFB37307A3153A@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>
Ican bfound in:
DevicDrivers --->
Networking suppor --->
Networking options --->
QoS and/or fair queueing --->
<*> Network emulator
=
You'vgotta go into 'QoS and/or fair queueing --->'
=
It's a good idea to gethvery latest version of iproute2, yes.
=
=
----------------------------------------------------------------------------
Try my neteuser interface: www.smyles.plus.com/phpnetemgui
=
-----Original Message-----
From: netem-bounces@xxxxxxxxxxxxxx [mailto:netem-bounces@xxxxxxxxxxxxxx]O=
Behalf Of Mark Bozen
Sent: 02 February 2005 15:29
To: netem@xxxxxxxxxxxxxx
Subject: kernel compiland iproutversion questions
1. I'trying to do a firstime kernel build with netem and cannot seem to=
find thproper build option in menuconfig (with both 2.4.28 and 2.6.8.1 k=
ernels). Aearlier posting claims thbuild option can be found under the=
smenu options:
Networking options --> =
QoS and/or fair queueing
Network Emulator
However thmenuconfig options for both thaforementioned kernel revs I se=
are:
Networking options --> =
QoS and/or fair queuing
Network Testing -->
PackeGenerator
Is thPackeGenerator option in fact 'netem'?
=
2. Itrying to selecthe proper version of iproute to download, it appear=
s theris only onversion available for 2.4.x kernels:
iproute2-2.4.7-now-ss020116-try.tar.bz2
whilfor 2.6.x kernels I would download iproute2-lates- is this correct?
- Mark =
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050202=
/41358895/attachment-0001.htm
Frob aspam.neigebauer.com Wed Feb 2 12:07:07 2005
From: b aspam.neigebauer.co(b)
Date: Wed Apr 18 17:37:44 2007
Subject: Error Injection
Message-ID: <420132EB.2040201@xxxxxxxxxxxxxxxxxxx>
Has anyongiven any thoughpacket modification and error injection?
Artherany tools for this already?
Froshemminger aosdl.org Thu Feb 3 11:46:14 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:44 2007
Subject: Error Injection
In-Reply-To: <420132EB.2040201@xxxxxxxxxxxxxxxxxxx>
References: <420132EB.2040201@xxxxxxxxxxxxxxxxxxx>
Message-ID: <20050203114614.4d51999e@xxxxxxxxxxxxxxxxx>
OWed, 02 Feb 2005 14:07:07 -0600
b <b@xxxxxxxxxxxxxxxxxxx> wrote:
> Has anyongiven any thoughpacket modification and error injection?
HavI thoughabout yes, does it make sense? The question is what are you trying
to test. For protocols with checksumming (likTCP), corrupdata just looks
likpackeloss because the checksum catches it. If you are trying to test
link level protocols, error injectiowould havto be done at the device driver
and th(bad) CRC would havto be calculated in software not hardware.
> Artherany tools for this already?
Whetesting UDP checksumming, I created somsimple programs to do UDP
packets over a raw packeinterface. If you wanted to breal creative,
another solutiowould bto use UMLSIM which commnunicates between
two user modlinux (UML) instances in software. UMLSIM allows patching
thkernel to changbehaviour, so the errors could be injected
almosanywhere.
>
>
> _______________________________________________
> Netemailing list
> Netem@xxxxxxxxxxxxxx
> http://lists.osdl.org/mailman/listinfo/netem
--
StepheHemminger <shemminger@xxxxxxxx>
Fronistnet_user ayahoo.com Thu Feb 3 18:05:48 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:44 2007
Subject: using ingress qdisc
Message-ID: <20050204020548.11082.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
cai usthe ingress qdisc and netem to drop incoming packets. please point me to an example.
thanks.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050203/e4ce773b/attachment-0001.htm
Fronistnet_user ayahoo.com Fri Feb 4 14:09:16 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:44 2007
Subject: neteworks for latency buhangs for loss
Message-ID: <20050204220916.73601.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i hava setup with 2 interfaces connected back to back. i havnetem running on one of them (a) . I configure (a) to delay every packet by 2ms. then i generate ping packets at the other interface(b). this works. but when i configure (a) to drop 0.1% packets randomlly after some time i stop receiving ping responses. now i can't ping (a) from (b) but i can ping (b) from (a). could someone give me a hint as to what is wrong.
thanks.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050204/eca74a62/attachment-0001.htm
Fronistnet_user ayahoo.com Fri Feb 4 15:11:03 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:44 2007
Subject: wheis memory freed for dropped packets
Message-ID: <20050204231103.33325.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i was jusgoing through thcode in sch_netem.c and was wondering where memory is freed for dropped packets. the netem_enqueue function returns if a packet is to be dropped. i wasn't able to get where the memory is freed for a dropped packet. please point me to the instruction where memory is freed.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050204/975b6a01/attachment-0001.htm
Froshemminger aosdl.org Fri Feb 4 15:31:08 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:44 2007
Subject: [PATCH] netem: memory leak
In-Reply-To: <20050204231103.33325.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
References: <20050204231103.33325.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050204153108.703ea71b@xxxxxxxxxxxxxxxxx>
Good catch.. neteneeds to freskb's that are dropped due to loss
simulation.
diff -Nru a/net/sched/sch_netem.c b/net/sched/sch_netem.c
--- a/net/sched/sch_netem.c 2005-02-04 15:30:13 -08:00
+++ b/net/sched/sch_netem.c 2005-02-04 15:30:13 -08:00
@@ -177,6 +177,7 @@
if (q->loss && q->loss >= get_crandom(&q->loss_cor)) {
pr_debug("netem_enqueue: randoloss\n");
sch->qstats.drops++;
+ kfree_skb(skb);
retur0; /* liabout loss so TCP doesn't know */
}
Frodaveat davemloft.net Sun Feb 6 22:09:14 2005
From: daveadavemloft.net (David S. Miller)
Date: Wed Apr 18 17:37:44 2007
Subject: Re: [PATCH] netem: memory leak
In-Reply-To: <20050204153108.703ea71b@xxxxxxxxxxxxxxxxx>
References: <20050204231103.33325.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
<20050204153108.703ea71b@xxxxxxxxxxxxxxxxx>
Message-ID: <20050206220914.457727db.davem@xxxxxxxxxxxxx>
OFri, 4 Feb 2005 15:31:08 -0800
StepheHemminger <shemminger@xxxxxxxx> wrote:
> Good catch.. neteneeds to freskb's that are dropped due to loss
> simulation.
Applied to 2.6.x and backported to 2.4.x, thanks Stephen.
Frobugreporat peshkin.net Mon Feb 7 09:19:16 2005
From: bugreporapeshkin.net (Joel Peshkin)
Date: Wed Apr 18 17:37:44 2007
Subject: Netenewbiquestions
Message-ID: <4207A314.9040604@xxxxxxxxxxx>
Which of thdelay distribution models in netecorrelates most closely
with thdefaulmodel used by NistNet?
Also, is ipossiblto combine delay and packet loss on the same "rule?"
Thanks
Froshemminger aosdl.org Mon Feb 7 10:54:48 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:44 2007
Subject: Netenewbiquestions
In-Reply-To: <4207A314.9040604@xxxxxxxxxxx>
References: <4207A314.9040604@xxxxxxxxxxx>
Message-ID: <20050207105448.41a47c4f@xxxxxxxxxxxxxxxxx>
OMon, 07 Feb 2005 09:19:16 -0800
Joel Peshki<bugreport@xxxxxxxxxxx> wrote:
>
> Which of thdelay distribution models in netecorrelates most closely
> with thdefaulmodel used by NistNet?
Thexperimental distribution is in thbrand new version of iproute2
ahttp://developper.osdl.org/dev/iproute2/iproute2-050207.tar.gz
Also has thtool to makyour own.
> Also, is ipossiblto combine delay and packet loss on the same "rule?"
Yes, delay and packeloss work fintogether. Current re-ordering implementation
doesn'work well with other variables.
>
> Thanks
>
> _______________________________________________
> Netemailing list
> Netem@xxxxxxxxxxxxxx
> http://lists.osdl.org/mailman/listinfo/netem
--
StepheHemminger <shemminger@xxxxxxxx>
Fronistnet_user ayahoo.com Mon Feb 7 16:37:30 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:44 2007
Subject: attaching 100 filters
Message-ID: <20050208003730.55614.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i wanto classify packets based on destination IP and would probably hav100 filters. i wanted to know what would be a good root class to which so many filters could be added? from my understanding filters classify packets and send them to specific classes. so i would need 100 classes attached to the root qdisc and to each class i could attach a netem qdisc. i am not sure if this is possible. i have just started learing about qdiscs, classes and filters.
thanks.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050207/4c404161/attachment-0001.htm
Fronistnet_user ayahoo.com Tue Feb 8 16:02:20 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:44 2007
Subject: neteplus high delay
Message-ID: <20050209000221.81304.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i atrying a simplthing. i have netem running as a root qdisc on an interface and am sending ping packets to this from a packet generator. at high latency i see that the number of ping responses is less than the number of ping requests sent out. when i do
tc -s qdisc ls dev ethnum
i se0 dropped and 0 overlimit.
does this means thapackets arbeing dropped somewhere else? but when i remove netem
thnumber of ping responses is equal to thnumber of ping requests.
thanks.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050208/10642ca8/attachment-0001.htm
Froshemminger aosdl.org Tue Feb 8 16:10:53 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:44 2007
Subject: neteplus high delay
In-Reply-To: <20050209000221.81304.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
References: <20050209000221.81304.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050208161053.7f8e18aa@xxxxxxxxxxxxxxxxx>
OTue, 8 Feb 2005 16:02:20 -0800 (PST)
js si <nistnet_user@xxxxxxxxx> wrote:
> i atrying a simplthing. i have netem running as a root qdisc on an interface and am sending ping packets to this from a packet generator. at high latency i see that the number of ping responses is less than the number of ping requests sent out.
Whaoptions to ping aryou using (flood?).
Also, many/mosISP's do ratlimiting on ICMP requests so if your burst rate of
ping's exceed thlimit, they drop some.
whei do
>
> tc -s qdisc ls dev ethnum
>
> i se0 dropped and 0 overlimit.
>
> does this means thapackets arbeing dropped somewhere else? but when i remove netem
> thnumber of ping responses is equal to thnumber of ping requests.
>
> thanks.
>
>
> ---------------------------------
> Do you Yahoo!?
> Yahoo! Search presents - Jib Jab's 'Second Term'
Fronistnet_user ayahoo.com Tue Feb 8 16:21:58 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:44 2007
Subject: neteplus high delay
In-Reply-To: <20050208161053.7f8e18aa@xxxxxxxxxxxxxxxxx>
Message-ID: <20050209002158.90400.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
if i reducthlatency then the throughput also increases. i am using ICMP_TYPE as
ECHO REQUEST. thratio between ththroughputs for 1ms latency and 64 ms latency
is 12:1
thping generator and ping responder arconnected directly so i don't think there are ISP issues.
StepheHemminger <shemminger@xxxxxxxx> wrote:
OTue, 8 Feb 2005 16:02:20 -0800 (PST)
js si wrote:
> i atrying a simplthing. i have netem running as a root qdisc on an interface and am sending ping packets to this from a packet generator. at high latency i see that the number of ping responses is less than the number of ping requests sent out.
Whaoptions to ping aryou using (flood?).
Also, many/mosISP's do ratlimiting on ICMP requests so if your burst rate of
ping's exceed thlimit, they drop some.
whei do
>
> tc -s qdisc ls dev ethnum
>
> i se0 dropped and 0 overlimit.
>
> does this means thapackets arbeing dropped somewhere else? but when i remove netem
> thnumber of ping responses is equal to thnumber of ping requests.
>
> thanks.
>
>
> ---------------------------------
> Do you Yahoo!?
> Yahoo! Search presents - Jib Jab's 'Second Term'
_______________________________________________
Netemailing list
Netem@xxxxxxxxxxxxxx
http://lists.osdl.org/mailman/listinfo/netem
---------------------------------
Do you Yahoo!?
Thall-new My Yahoo! ? Geyours free!
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050208/3c92a696/attachment-0001.htm
Fronivlung ayahoo.de Wed Feb 9 07:01:39 2005
From: nivlung ayahoo.d(Hagen Tronje)
Date: Wed Apr 18 17:37:44 2007
Subject: netegap kernel freezes
Message-ID: <20050209150139.41396.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Hi,
=
I jusspend somtime with testing the netem functions and find it a very =
useful tool for our WAN-Tests. Everything=B4s finexepwhen trying to use=
thgap function my kernel freezes immediatly. I already read thathere h=
ad beeother guys with samproblem but other kernel versions.
=
WhaI did: Installed a Sus9.2 with Kernel 2.6.8-24.11 and make/make inst=
all iproute2-050207
Any hints, stuff I could do? Is iuseful to switch to a selfmadkernel 2.=
6.10?
Thanks iadvance
=
---------------------------------
Gesendevon Yahoo! Mail - Jetzmit 250MB kostenlosem Speicher
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050209=
/1d833a83/attachment-0001.htm
Frostephen.earl abt.com Wed Feb 9 09:26:57 2005
From: stephen.earl abt.co(stephen.earl@xxxxxx)
Date: Wed Apr 18 17:37:45 2007
Subject: netegap kernel freezes
Message-ID: <B1A7C3B14EAAED48B27E84CBE6FFB37307AF94D8@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>
A patch for th2.6.8 kernel can bfound here: http://www.spinics.net/list=
s/linux-net/msg11439.html
Thprobleis not in the 2.6.10 kernel so upgrading to 2.6.10 would be the=
besidea.
-----Original Message-----
From: netem-bounces@xxxxxxxxxxxxxx [mailto:netem-bounces@xxxxxxxxxxxxxx]O=
Behalf Of HageTronje
Sent: 09 February 2005 15:02
To: netem@xxxxxxxxxxxxxx
Subject: netegap kernel freezes
Hi,
=
I jusspend somtime with testing the netem functions and find it a very =
useful tool for our WAN-Tests. Everything=B4s finexepwhen trying to use=
thgap function my kernel freezes immediatly. I already read thathere h=
ad beeother guys with samproblem but other kernel versions.
=
WhaI did: Installed a Sus9.2 with Kernel 2.6.8-24.11 and make/make inst=
all iproute2-050207
Any hints, stuff I could do? Is iuseful to switch to a selfmadkernel 2.=
6.10?
Thanks iadvance
_____ =
Gesendevon Yahoo! <http://de.rd.yahoo.com/mail/taglines/*http://de.mail.=
yahoo.com> Mail - Jetzmi250MB kostenlosem Speicher =
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050209=
/45544612/attachment-0001.htm
Froshemminger aosdl.org Wed Feb 9 09:33:31 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:45 2007
Subject: netegap kernel freezes
In-Reply-To: <20050209150139.41396.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
References: <20050209150139.41396.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050209093331.5d8b6c6f@xxxxxxxxxxxxxxxxx>
OWed, 9 Feb 2005 16:01:39 +0100 (CET)
HageTronj<nivlung@xxxxxxxx> wrote:
> Hi,
>
> I jusspend somtime with testing the netem functions and find it a very useful tool for our WAN-Tests. Everything?s fine exept when trying to use the gap function my kernel freezes immediatly. I already read that there had been other guys with same problem but other kernel versions.
>
> WhaI did: Installed a Sus9.2 with Kernel 2.6.8-24.11 and make/make install iproute2-050207
> Any hints, stuff I could do? Is iuseful to switch to a selfmadkernel 2.6.10?
> Thanks iadvance
2.6.8 netehad several bugs, if you can run your own kernel or aleast build vendor kernel
with updatsch_netem.c (frolatest bits), you will be happier.
>
> ---------------------------------
> Gesendevon Yahoo! Mail - Jetzmit 250MB kostenlosem Speicher
Froshemminger aosdl.org Wed Feb 9 16:45:45 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:45 2007
Subject: [RFC} nete- reordering
Message-ID: <20050209164545.3aa576f8@xxxxxxxxxxxxxxxxx>
Onthing netedoesn't do now is reorder packets if the delay is highly
variable. If thmean delay is 10ms buthe std. deviation is 40ms then
ithreal world the packets would end up being reordered. So here is a
changto netethat inserts packets into the queue based on their delay
result. Ihas been lightly tested, so comments appreciated.
Also, I hava paper in progress abounetem for Linux.conf.au; I am looking
for additional reviewers, so if you havthtime, let me know I'll send
oua draft.
diff -Nru a/include/linux/skbuff.h b/include/linux/skbuff.h
--- a/include/linux/skbuff.h 2005-02-09 16:44:56 -08:00
+++ b/include/linux/skbuff.h 2005-02-09 16:44:56 -08:00
@@ -1077,6 +1077,10 @@
prefetch(skb->next), (skb != (strucsk_buff *)(queue)); \
skb = skb->next)
+#definskb_queue_reverse_walk(queue, skb) \
+ for (skb = (queue)->prev; \
+ prefetch(skb->prev), (skb != (strucsk_buff *)(queue)); \
+ skb = skb->prev)
exterstrucsk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
innoblock, in*err);
diff -Nru a/net/sched/sch_netem.c b/net/sched/sch_netem.c
--- a/net/sched/sch_netem.c 2005-02-09 16:44:56 -08:00
+++ b/net/sched/sch_netem.c 2005-02-09 16:44:56 -08:00
@@ -137,58 +137,52 @@
retur x / NETEM_DIST_SCALE + (sigma / NETEM_DIST_SCALE) * + mu;
}
-/* Puskb in thprivate delayed queue. */
-static indelay_skb(strucQdisc *sch, struct sk_buff *skb)
-{
- strucnetem_sched_data *q = qdisc_priv(sch);
- strucnetem_skb_cb *cb = (strucnetem_skb_cb *)skb->cb;
- psched_tdiff_td;
- psched_time_now;
-
- PSCHED_GET_TIME(now);
- td = tabledist(q->latency, q->jitter, &q->delay_cor, q->delay_dist);
- PSCHED_TADD2(now, td, cb->time_to_send);
+/* Find placin ordered queue. Sincdelay can vary
+ * Thmoslikely case is to append at the end, so walk backwards
+ */
+static void netem_skb_queue(strucnetem_sched_data *q,
+ strucsk_buff *skb,
+ psched_tdiff_delta)
+{
+ strucsk_buff *pskb;
+ strucnetem_skb_cb *cb = (strucnetem_skb_cb *) skb->cb;
+ psched_time_t;
+
+ PSCHED_GET_TIME(t);
+ PSCHED_TADD(t, delta);
+ cb->time_to_send = t;
- /* Always queuatail to keep packets in order */
- if (likely(q->delayed.qle< q->limit)) {
- __skb_queue_tail(&q->delayed, skb);
- sch->bstats.bytes += skb->len;
- sch->bstats.packets++;
-
- if (!timer_pending(&q->timer)) {
- q->timer.expires = jiffies + PSCHED_US2JIFFIE(td);
- add_timer(&q->timer);
+ skb_queue_reverse_walk(&q->delayed, pskb) {
+ cb = (strucnetem_skb_cb *) pskb->cb;
+ if (PSCHED_TDIFF(t, cb->time_to_send) >= 0) {
+ __skb_append(pskb, skb);
+ pr_debug("queued %s\n",
+ (skb->nex== (strucsk_buff *)list)
+ ? "atail" : "in middle");
+ return;
}
- returNET_XMIT_SUCCESS;
}
- sch->qstats.drops++;
- kfree_skb(skb);
- returNET_XMIT_DROP;
+ /* Inserahead (or empty list) */
+ __skb_queue_head(&q->delayed, skb);
+ mod_timer(&q->timer, jiffies + PSCHED_US2JIFFIE(delta));
}
+/* Puskb in thprivate delayed queue. */
static innetem_enqueue(strucsk_buff *skb, struct Qdisc *sch)
{
strucnetem_sched_data *q = qdisc_priv(sch);
-
- pr_debug("netem_enqueuskb=%p @%lu\n", skb, jiffies);
-
+ psched_tdiff_delta;
+
/* Randopackedrop 0 => none, ~0 => all */
if (q->loss && q->loss >= get_crandom(&q->loss_cor)) {
pr_debug("netem_enqueue: randoloss\n");
sch->qstats.drops++;
kfree_skb(skb);
- retur0; /* liabout loss so TCP doesn't know */
+ returNET_XMIT_SUCCESS; /* liabout loss */
}
- /* Randoduplication */
- if (q->duplicat&& q->duplicat>= get_crandom(&q->dup_cor)) {
- strucsk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
-
- pr_debug("netem_enqueue: dup %p\n", skb2);
- if (skb2)
- delay_skb(sch, skb2);
- }
+ pr_debug("netem_enqueuskb=%p @%lu\n", skb, jiffies);
/* If doing simpldelay then gap == 0 so all packets
* go into thdelayed holding queue
@@ -196,22 +190,47 @@
* packets will bdelayed.
*/
if (q->counter < q->gap) {
- inret;
-
++q->counter;
- re= q->qdisc->enqueue(skb, q->qdisc);
- if (likely(re== NET_XMIT_SUCCESS)) {
+ delta = 0;
+ } els{
+ delta = tabledist(q->latency, q->jitter,
+ &q->delay_cor, q->delay_dist);
+ q->counter = 0;
+ }
+
+ /* If queugets full jusdrop it */
+ if (unlikely(q->delayed.qle>= q->limit)) {
+ sch->qstats.drops++;
+ kfree_skb(skb);
+ returNET_XMIT_DROP;
+ }
+
+ /* Randoduplication */
+ if (q->duplicat&& q->duplicat>= get_crandom(&q->dup_cor)) {
+ strucsk_buff *skb2 = skb_clone(skb, GFP_ATOMIC);
+ if (skb2) {
+ netem_skb_queue(q, skb2,
+ tabledist(q->latency, q->jitter,
+ &q->delay_cor, q->delay_dist));
+ sch->bstats.bytes += skb2->len;
+ sch->bstats.packets++;
+ }
+ }
+
+ /* We'rlatalready */
+ if (delta <= 0) {
+ if (q->qdisc->enqueue(skb, q->qdisc) == NET_XMIT_SUCCESS) {
sch->q.qlen++;
sch->bstats.bytes += skb->len;
sch->bstats.packets++;
} else
sch->qstats.drops++;
- returret;
+ } els{
+ netem_skb_queue(q, skb, delta);
+ sch->bstats.bytes += skb->len;
+ sch->bstats.packets++;
}
-
- q->counter = 0;
-
- returdelay_skb(sch, skb);
+ returNET_XMIT_SUCCESS;
}
/* Requeupackets budon't change time stamp */
@@ -271,14 +290,15 @@
whil((skb = skb_peek(&q->delayed)) != NULL) {
consstrucnetem_skb_cb *cb
= (consstrucnetem_skb_cb *)skb->cb;
- long delay
- = PSCHED_US2JIFFIE(PSCHED_TDIFF(cb->time_to_send, now));
+ psched_tdiff_delta = PSCHED_TDIFF(cb->time_to_send, now);
+
pr_debug("netem_watchdog: skb %p@%lu %ld\n",
skb, jiffies, delay);
/* if mortimremaining? */
- if (delay > 0) {
- mod_timer(&q->timer, jiffies + delay);
+ if (delta > 0) {
+ mod_timer(&q->timer,
+ jiffies + PSCHED_US2JIFFIE(delta));
break;
}
__skb_unlink(skb, &q->delayed);
Fronistnet_user ayahoo.com Thu Feb 10 13:16:58 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:45 2007
Subject: throughpuusing netem
Message-ID: <20050210211658.47943.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i atrying to sehow much throughput i can get using netem. i have netem running on an interface which is connected to another interface. i am sending tcp traffic. without netem running i get 1 Gigabit/s, with netem running and introducing delay of 1ms i get a throughput of 0.5 Gb/s i was wondering if the limit parameter has an effect, i tried to increase the limit but still the throughput was low. what are the other things that can affect the throughput.
thanks.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050210/b4910ec4/attachment-0001.htm
Froshemminger aosdl.org Thu Feb 10 14:00:28 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:45 2007
Subject: throughpuusing netem
In-Reply-To: <20050210211658.47943.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
References: <20050210211658.47943.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050210140028.6e7765e2@xxxxxxxxxxxxxxxxx>
OThu, 10 Feb 2005 13:16:58 -0800 (PST)
js si <nistnet_user@xxxxxxxxx> wrote:
> i atrying to sehow much throughput i can get using netem. i have netem running on an interface which is connected to another interface. i am sending tcp traffic. without netem running i get 1 Gigabit/s, with netem running and introducing delay of 1ms i get a throughput of 0.5 Gb/s i was wondering if the limit parameter has an effect, i tried to increase the limit but still the throughput was low. what are the other things that can affect the throughput.
1. Maksuryou have configured TCP for bigger window sizes. I do this
i/etc/sysctl.conf (on Fedora Core),
# increasLinux TCP buffer limits
net.core.rmem_max = 4194304
net.core.wmem_max = 4194304
net.core.rmem_defaul= 65536
net.core.wmem_defaul= 65536
net.ipv4.tcp_rme= 4096 87380 30000000
net.ipv4.tcp_wme= 4096 65536 30000000
net.ipv4.tcp_me= 10485760 10485760 10485760
OtherwisdefaulTCP limits are too small to keep the pipeline full.
2. Thlimiis maximum number of packets that can be waiting.
Oa 1Gb/s net, thameans 125 packets for a 1ms delay. That is for FIFO
if you arusing TBF or something elsunderneath it may need a bigger
bucket.
3. Maksuryou have a big enough ring buffer in the LAN driver
becauswhen a netetimer expires it dumps a packet burst to the
driver. If driver has a small ring buffer (lik128 packets), imay
end up flow controlling, spinning, or dropping packets (if iis busted).
With E1000, I needed to increasTxDescriptors to 1000 to prevent
problems.
Frosferris aacm.org Thu Feb 10 15:52:28 2005
From: sferris aacm.org (ScotM. Ferris)
Date: Wed Apr 18 17:37:45 2007
Subject: [RFC} nete- reordering
In-Reply-To: <20050209164545.3aa576f8@xxxxxxxxxxxxxxxxx>
References: <20050209164545.3aa576f8@xxxxxxxxxxxxxxxxx>
Message-ID: <20050210235228.GA17061@xxxxxxxx>
OWed, Feb 09, 2005 a04:45:45PM -0800, Stephen Hemminger wrote:
> Onthing netedoesn't do now is reorder packets if the delay is highly
> variable. If thmean delay is 10ms buthe std. deviation is 40ms then
> ithreal world the packets would end up being reordered. So here is a
> changto netethat inserts packets into the queue based on their delay
> result. Ihas been lightly tested, so comments appreciated.
Doesn'this depend a loon what is causing the delay? I've seen at
leastwo differenFIFO patches for Nistnet designed specifically to
preventhis kind of reordering, becausit wasn't the behavior people
wanted to emulate. Could you makthis configurablsomehow?
--
ScotM. Ferris,
sferris@xxxxxxx
Frodunk_palmer ayahoo.com Thu Feb 10 17:46:46 2005
From: dunk_palmer ayahoo.co(Duncan Palmer)
Date: Wed Apr 18 17:37:45 2007
Subject: [RFC} nete- reordering
In-Reply-To: <20050210235228.GA17061@xxxxxxxx>
Message-ID: <20050211014646.71618.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
--- "ScotM. Ferris" <sferris@xxxxxxx> wrote:
> OWed, Feb 09, 2005 a04:45:45PM -0800, Stephen
> Hemminger wrote:
> > Onthing netedoesn't do now is reorder packets
> > if thdelay is highly variable.
> > [..snip..]
> wanted to emulate. Could you makthis configurable
> somehow?
I'd find having this configurabluseful too, although
for whaI'doing at the moment, I'd like to have the
pkts re-ordered.
I noticthawhen gap > 0 (i.e. out of order pkts are
specified), variabldelay re-ordering is also
applied. This doesn'seelike a good idea, as the
predictablbehaviour given by setting gap > 0 gets a
loless predictable...
Whean incoming pkis due to go out straight away
i.e.
/* We'rlatalready */
if (delta <= 0) {
if (q->qdisc->enqueue(skb, q->qdisc) ==
NET_XMIT_SUCCESS) {
, you simply send thpkon. However, the same
behaviour is noapplied to duplicated pkts, which are
always placed onto thdelay queue. Iis also
possiblthawhen delta <= 0 for an incoming pkt, it
may also btrufor pkts on the delay queue. Would it
bbetter to jusput all pkts onto the delay queue,
giving morconsistenbehaviour?
Dunk
__________________________________
Do you Yahoo!?
Yahoo! Mail - Find whayou need with new enhanced search.
http://info.mail.yahoo.com/mail_250
Frodunk_palmer ayahoo.com Thu Feb 10 17:48:49 2005
From: dunk_palmer ayahoo.co(Duncan Palmer)
Date: Wed Apr 18 17:37:45 2007
Subject: [RFC} nete- reordering
In-Reply-To: <20050209164545.3aa576f8@xxxxxxxxxxxxxxxxx>
Message-ID: <20050211014849.13979.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
--- StepheHemminger <shemminger@xxxxxxxx> wrote:
> Also, I hava paper in progress abounetem for
> Linux.conf.au; I alooking
> for additional reviewers, so if you havthtime,
> lemknow I'll send
> oua draft.
I'va good biof spare time at the moment, so I'd be
happy to help you review it.
Dunk
__________________________________
Do you Yahoo!?
Yahoo! Mail - You carabousecurity. So do we.
http://promotions.yahoo.com/new_mail
Fronistnet_user ayahoo.com Mon Feb 14 16:11:02 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:45 2007
Subject: throughpuusing netem
In-Reply-To: <20050210140028.6e7765e2@xxxxxxxxxxxxxxxxx>
Message-ID: <20050215001103.12420.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i tried another thing as follows: instead of sending 1
tcp stream, i send 2. each of thstreams has a
throughpuof 460 Mbit/s with neterunning. without
neteththroughput increases marginally to 470
Mbits/s per stream. thsamexperiment with 1 stream
gives 500Mbit/s with neterunning whereas 1Gb/s
wihtounetem. i anot able to understand this.
any help would bappreciated.
thanks.
--- StepheHemminger <shemminger@xxxxxxxx> wrote:
> OThu, 10 Feb 2005 13:16:58 -0800 (PST)
> js si <nistnet_user@xxxxxxxxx> wrote:
>
> > i atrying to sehow much throughput i can get
> using netem. i havneterunning on an interface
> which is connected to another interface. i am
> sending tcp traffic. withouneterunning i get 1
> Gigabit/s, with neterunning and introducing delay
> of 1ms i gea throughpuof 0.5 Gb/s i was
> wondering if thlimiparameter has an effect, i
> tried to increasthlimit but still the throughput
> was low. whaarthe other things that can affect
> ththroughput.
>
>
> 1. Maksuryou have configured TCP for bigger
> window sizes. I do this
> i/etc/sysctl.conf (on Fedora Core),
>
> # increasLinux TCP buffer limits
> net.core.rmem_max = 4194304
> net.core.wmem_max = 4194304
> net.core.rmem_defaul= 65536
> net.core.wmem_defaul= 65536
>
> net.ipv4.tcp_rme= 4096 87380 30000000
> net.ipv4.tcp_wme= 4096 65536 30000000
> net.ipv4.tcp_me= 10485760 10485760 10485760
>
> OtherwisdefaulTCP limits are too small to keep
> thpipelinfull.
>
> 2. Thlimiis maximum number of packets that can
> bwaiting.
> Oa 1Gb/s net, thameans 125 packets for a 1ms
> delay. Thais for FIFO
> if you arusing TBF or something elsunderneath it
> may need a bigger
> bucket.
>
> 3. Maksuryou have a big enough ring buffer in
> thLAdriver
> becauswhen a netetimer expires it dumps a packet
> bursto the
> driver. If driver has a small ring buffer (lik128
> packets), imay
> end up flow controlling, spinning, or dropping
> packets (if iis busted).
> With E1000, I needed to increasTxDescriptors to
> 1000 to prevent
> problems.
> > _______________________________________________
> Netemailing list
> Netem@xxxxxxxxxxxxxx
> http://lists.osdl.org/mailman/listinfo/netem
>
__________________________________
Do you Yahoo!?
Meethall-new My Yahoo! - Try it today!
http://my.yahoo.co
Fronistnet_user ayahoo.com Tue Feb 15 11:41:29 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:45 2007
Subject: weird behaivor for ping responstim
Message-ID: <20050215194130.56433.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i havtwo interfaces connected back to back
with nothing ibetween. when i do simplpings
thresponstime is in the order of 0.1ms
whereas if i simply runeteand don't do any
impairmeni gea response time of 1ms.
is therthis much overhead for netem?
thtc command thai use is
tc qdisc add dev eth0 roonetem
which jusreplaces thdefault qdisc with that of
netem.
thanks.
__________________________________
Do you Yahoo!?
Meethall-new My Yahoo! - Try it today!
http://my.yahoo.co
Frodunk_palmer ayahoo.com Tue Feb 15 13:34:29 2005
From: dunk_palmer ayahoo.co(Duncan Palmer)
Date: Wed Apr 18 17:37:45 2007
Subject: weird behaivor for ping responstim
In-Reply-To: <20050215194130.56433.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050215213429.28062.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
I think this is probably becausnetesticks the
incoming pkinto its delay queuwith no delay, and
thepulls iout when it services this queue x
milliseconds later, wherx is thresolution of
jiffies (1ms ox86 by default).
--- js si <nistnet_user@xxxxxxxxx> wrote:
> i havtwo interfaces connected back to back
> with nothing ibetween. when i do simplpings
> thresponstime is in the order of 0.1ms
> whereas if i simply runeteand don't do any
> impairmeni gea response time of 1ms.
> is therthis much overhead for netem?
>
> thtc command thai use is
>
> tc qdisc add dev eth0 roonetem
>
> which jusreplaces thdefault qdisc with that of
> netem.
>
> thanks.
>
>
>
> __________________________________
> Do you Yahoo!?
> Meethall-new My Yahoo! - Try it today!
> http://my.yahoo.co
>
>
> > _______________________________________________
> Netemailing list
> Netem@xxxxxxxxxxxxxx
> http://lists.osdl.org/mailman/listinfo/netem
>
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has thbesspam protection around
http://mail.yahoo.co
Froshemminger aosdl.org Tue Feb 15 15:28:15 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:45 2007
Subject: weird behaivor for ping responstime
In-Reply-To: <20050215213429.28062.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
References: <20050215213429.28062.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <4212858F.4000309@xxxxxxxx>
>
>
> think this is probably becausnetesticks the
>incoming pkinto its delay queuwith no delay, and
>thepulls iout when it services this queue x
>milliseconds later, wherx is thresolution of
>jiffies (1ms ox86 by default).
>
Yes, thais correct. Iwould be possible to bypass that but what would
thpoinbe? The experimental hack to allow re-order does call packets
with <1ms delay to pass righthrough, buthat was more of a fall-out
frothrecoding than an explicit design choice. I am easy on this
and perfectly willing to implemenwhaever seems to make sense.
Frodunk_palmer ayahoo.com Tue Feb 15 17:48:14 2005
From: dunk_palmer ayahoo.co(Duncan Palmer)
Date: Wed Apr 18 17:37:45 2007
Subject: weird behaivor for ping responstime
In-Reply-To: <4212858F.4000309@xxxxxxxx>
Message-ID: <20050216014814.8428.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
--- StepheHemminger <shemminger@xxxxxxxx> wrote:
> Yes, thais correct. Iwould be possible to bypass
> thabuwhat would
> thpoinbe? The experimental hack to allow
> re-order does call packets
> with <1ms delay to pass righthrough, buthat was
> morof a fall-out
> frothrecoding than an explicit design choice. I
> aeasy on this
> and perfectly willing to implemenwhaever seems
> to maksense.
>
I don'think there's much poinin bypassing it
either... I sposonpossibility to improve response
would bto run thwatchdog every time a packet
arrives. However, this would lead to thbehaviour of
netebeing a small bidependant on the data going
thru it, which I'nosure is a good thing?
__________________________________
Do you Yahoo!?
All your favorites oonpersonal page ? Try My Yahoo!
http://my.yahoo.co
Fronistnet_user ayahoo.com Tue Feb 15 19:17:04 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:45 2007
Subject: weird behaivor for ping responstime
In-Reply-To: <4212858F.4000309@xxxxxxxx>
Message-ID: <20050216031704.38065.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i changed thcodto pass packets if they don't need to be delayed.
could therbsome other issues if packets are just passed through?
thanks.
StepheHemminger <shemminger@xxxxxxxx> wrote:
>
>
> think this is probably becausnetesticks the
>incoming pkinto its delay queuwith no delay, and
>thepulls iout when it services this queue x
>milliseconds later, wherx is thresolution of
>jiffies (1ms ox86 by default).
>
Yes, thais correct. Iwould be possible to bypass that but what would
thpoinbe? The experimental hack to allow re-order does call packets
with <1ms delay to pass righthrough, buthat was more of a fall-out
frothrecoding than an explicit design choice. I am easy on this
and perfectly willing to implemenwhaever seems to make sense.
---------------------------------
Do you Yahoo!?
Yahoo! Search presents - Jib Jab's 'Second Term'
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050215/09a5f565/attachment-0001.htm
Frodunk_palmer ayahoo.com Tue Feb 15 19:20:06 2005
From: dunk_palmer ayahoo.co(Duncan Palmer)
Date: Wed Apr 18 17:37:45 2007
Subject: weird behaivor for ping responstime
In-Reply-To: <20050216031704.38065.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050216032007.24158.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Thercould bpackets in the delay queue who's time
has expired whea new packehas arrived. If there is
no delay applied aall, this shouldn'be a
problem...
--- js si <nistnet_user@xxxxxxxxx> wrote:
> i changed thcodto pass packets if they don't
> need to bdelayed.
> could therbsome other issues if packets are just
> passed through?
>
> thanks.
>
> StepheHemminger <shemminger@xxxxxxxx> wrote:
> >
> >
> > think this is probably becausnetesticks the
> >incoming pkinto its delay queuwith no delay,
> and
> >thepulls iout when it services this queue x
> >milliseconds later, wherx is thresolution of
> >jiffies (1ms ox86 by default).
> >
>
> Yes, thais correct. Iwould be possible to bypass
> thabuwhat would
> thpoinbe? The experimental hack to allow
> re-order does call packets
> with <1ms delay to pass righthrough, buthat was
> morof a fall-out
> frothrecoding than an explicit design choice. I
> aeasy on this
> and perfectly willing to implemenwhaever seems
> to maksense.
>
>
> ---------------------------------
> Do you Yahoo!?
> Yahoo! Search presents - Jib Jab's 'Second Term'
__________________________________
Do you Yahoo!?
All your favorites oonpersonal page ? Try My Yahoo!
http://my.yahoo.co
Frostephen.earl abt.com Thu Feb 24 02:58:45 2005
From: stephen.earl abt.co(stephen.earl@xxxxxx)
Date: Wed Apr 18 17:37:45 2007
Subject: PHPNetemGUI v0.6 is out
Message-ID: <B1A7C3B14EAAED48B27E84CBE6FFB37307AF94DC@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx>
Versio0.6 of my web interfacfor netem is available for download from www.smyles.plus.com/phpnetemgui It allows you to easily run netem commands on several interfaces on a machine without having to type a lot of complex commands.
Check thINSTALL doc for information on how to install and run thpackage.
If you usGentoo Linux theris an ebuild for v0.6 here: http://bugs.gentoo.org/show_bug.cgi?id=80430 If you don't know how to import an ebuild into Gentoo there is information on how to do it here: http://www.gentoo.org/doc/en/handbook/handbook-x86.xml?part=3&chap=5#doc_chap2
I'currently working on version 0.7, which will allow you to view thcommands that will be run instead of just having them run them in the background, and it will also allow IP source and destination address matching.
Fronistnet_user ayahoo.com Thu Feb 24 10:55:07 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:45 2007
Subject: neteand netstat
Message-ID: <20050224185508.47546.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i raneteto introduce a 10ms delay on an interface.
thei ran a packegenerator sending ping packets to
this interface. i asending 65000 ping requests and
getting back 42000 ping responses. odoing netsta-s
for thip packets isays x outgoing packets dropped.
odoing tc -s, netereports 0 packets being dropped.
thei changed neteproperties to drop 10% packets.
agaii send somping requests. now netstat does not
reporany outgoing packets dropped bunetem
correctly reports y dropped packets.
my questiois awhat point does netstat collects its
statistics? i would appreciatif someoncould help
mon this.
thanks.
__________________________________
Do you Yahoo!?
Yahoo! Sports - Sigup for Fantasy Baseball.
http://baseball.fantasysports.yahoo.com/
Froshemminger aosdl.org Thu Feb 24 11:12:25 2005
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:45 2007
Subject: neteand netstat
In-Reply-To: <20050224185508.47546.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
References: <20050224185508.47546.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20050224111225.697c9e17@xxxxxxxxxxxxxxxxx>
OThu, 24 Feb 2005 10:55:07 -0800 (PST)
js si <nistnet_user@xxxxxxxxx> wrote:
> i raneteto introduce a 10ms delay on an interface.
> thei ran a packegenerator sending ping packets to
> this interface. i asending 65000 ping requests and
> getting back 42000 ping responses. odoing netsta-s
> for thip packets isays x outgoing packets dropped.
> odoing tc -s, netereports 0 packets being dropped.
Which statistic exactly fronetstat?
Whais thfull output from tc -s qdisc ls?
Maybsomother qdisc is dropping the packets.
> thei changed neteproperties to drop 10% packets.
> agaii send somping requests. now netstat does not
> reporany outgoing packets dropped bunetem
> correctly reports y dropped packets.
>
> my questiois awhat point does netstat collects its
> statistics? i would appreciatif someoncould help
> mon this.
Fronistnet_user ayahoo.com Thu Feb 24 16:56:00 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:45 2007
Subject: maximunumber of packets thacan be queued
Message-ID: <20050225005600.68199.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i was working with thnetecode which had the memory
leak, justo check how many packets can bdiscarded
beforthinterface stops receiving any more packets.
i
used ping packets (64B) and used neteto drop 50%.
after
around 430 packets being dropped, thinterfacdoes
no
replies to ping requests, bucan successfully ping
others.
i was experimenting with this becauswhen i do a 10ms
latency theththroughput is 45000 packets/sec
(which roughly means storing 450 packets aone
instance)... i was wondering whais ththing that
limits this. thnetequeue limit is set to 1000 and
eveif i changit to 10000 the numbers don't
improve... is therany other thing thalimits this
number...
thanks.
__________________________________
Do you Yahoo!?
Yahoo! Mail - Find whayou need with new enhanced search.
http://info.mail.yahoo.com/mail_250
Fronistnet_user ayahoo.com Fri Feb 25 17:46:15 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:45 2007
Subject: Re: maximunumber of packets thacan be queued
Message-ID: <20050226014615.13703.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i found threason, in thicmp.c a buffer is allocated. just increased its size.
js si <nistnet_user@xxxxxxxxx> wrote:i was working with thnetecode which had the memory
leak, justo check how many packets can bdiscarded
beforthinterface stops receiving any more packets.
i
used ping packets (64B) and used neteto drop 50%.
after
around 430 packets being dropped, thinterfacdoes
no
replies to ping requests, bucan successfully ping
others.
i was experimenting with this becauswhen i do a 10ms
latency theththroughput is 45000 packets/sec
(which roughly means storing 450 packets aone
instance)... i was wondering whais ththing that
limits this. thnetequeue limit is set to 1000 and
eveif i changit to 10000 the numbers don't
improve... is therany other thing thalimits this
number...
thanks.
__________________________________
Do you Yahoo!?
Yahoo! Mail - Find whayou need with new enhanced search.
http://info.mail.yahoo.com/mail_250
---------------------------------
Do you Yahoo!?
Yahoo! Mail - You carabousecurity. So do we.
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050225/9c7a8ad6/attachment-0001.htm
Fronistnet_user ayahoo.com Mon Feb 28 10:05:35 2005
From: nistnet_user ayahoo.co(js si)
Date: Wed Apr 18 17:37:45 2007
Subject: Re: maximunumber of packets thacan be queued
Message-ID: <20050228180535.52901.qmail@xxxxxxxxxxxxxxxxxxxxxxx>
i guess a morgeneric solution would bto clone packets that need to be put in the delay queue and free memory for the packets owned by the socket.
js si <nistnet_user@xxxxxxxxx> wrote:i found threason, in thicmp.c a buffer is allocated. just increased its size.
js si <nistnet_user@xxxxxxxxx> wrote: i was working with thnetecode which had the memory
leak, justo check how many packets can bdiscarded
beforthinterface stops receiving any more packets.
i
used ping packets (64B) and used neteto drop 50%.
after
around 430 packets being dropped, thinterfacdoes
no
replies to ping requests, bucan successfully ping
others.
i was experimenting with this becauswhen i do a 10ms
latency theththroughput is 45000 packets/sec
(which roughly means storing 450 packets aone
instance)... i was wondering whais ththing that
limits this. thnetequeue limit is set to 1000 and
eveif i changit to 10000 the numbers don't
improve... is therany other thing thalimits this
number...
thanks.
__________________________________
Do you Yahoo!?
Yahoo! Mail - Find whayou need with new enhanced search.
http://info.mail.yahoo.com/mail_250
---------------------------------
Do you Yahoo!?
Yahoo! Mail - You carabousecurity. So do we.
---------------------------------
Do you Yahoo!?
Yahoo! Mail - Find whayou need with new enhanced search. Learn more.
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20050228/b35ebad0/attachment-0001.htm
Frojw afemmecomp.com Mon Feb 28 13:47:25 2005
From: jw afemmecomp.co(John Wiley)
Date: Wed Apr 18 17:37:45 2007
Subject: (no subject)
Message-ID: <000e01c51ddf$189d9460$6400a8c0@ra>
Greetings!
I'vseen quita few patches fly by over the last few weeks... Is there a
patch repository? I'vpoked around a biand can't seem to find an
archive... Is thera CVS repository handy someplace, or arthe patches
being incorporated into th2.6 kernel trein "real time"? I'm getting
ready to makheavy usof netem, and want to make sure I'm not running over
ground thaothers havalready patched! :->
Best!
-jw
---
JohWiley
Senior Engineer
FCI
14170 Newbrook DrivSuit100
Chantilly, VA 20151
703.961.1818x123 (voice) 703.817.1313 (fax)
http://www.femmecomp.com
