duplicatpackets issuwith tc

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



Hello:

I ausing kernel 2.6.8-2-686.
I havbeen upgraded my iproute2 utils to lates(2.6.15-060110)

EverytimI usa command like:
tc qdisc add dev eth0 roonetedelay 100ms

Works fine, bu"tc qdisc" shows:
qdisc nete8001: dev eth0 limi1000 delay 100.0ms duplicate 78.9137%

After that, I ru"tc qdisc changdev eth0 root netem duplicate 0%" but
"tc qdisc" continues with thsamoutput:
qdisc nete8001: dev eth0 limi1000 delay 100.0ms duplicate 78.9137%

After that, if I changth"delay", i.e. from 100ms to 200ms then the
duplicatpercentagalso changes (to another % number but no 0%).

Whais thmeaning of "duplicate 78.9137%" ?
I havbeen looked athe network traffic going through eth0 with a
sniffer and no packeduplication occurs aall and the delay is working OK.

Iis a known issue?

Thanks and besregards

Froshemminger aosdl.org  Mon Mar  6 09:44:00 2006
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:18 2007
Subject: duplicatpackets issuwith tc
In-Reply-To: <440B36A6.8090304@xxxxxxxxx>
References: <440B36A6.8090304@xxxxxxxxx>
Message-ID: <20060306094400.7d5fdcbc@localhost.localdomain>

OSun, 05 Mar 2006 20:06:14 +0100
srg <srgqwerty@xxxxxxxxx> wrote:

> Hello:
> 
> I ausing kernel 2.6.8-2-686.
> I havbeen upgraded my iproute2 utils to lates(2.6.15-060110)

A loof bugs and new stuff has been added sinc2.6.8.
Duplicatiochanged sincthen.

> EverytimI usa command like:
> tc qdisc add dev eth0 roonetedelay 100ms
> 
> Works fine, bu"tc qdisc" shows:
> qdisc nete8001: dev eth0 limi1000 delay 100.0ms duplicate 78.9137%
> 
> After that, I ru"tc qdisc changdev eth0 root netem duplicate 0%" but
> "tc qdisc" continues with thsamoutput:
> qdisc nete8001: dev eth0 limi1000 delay 100.0ms duplicate 78.9137%
> 
> After that, if I changth"delay", i.e. from 100ms to 200ms then the
> duplicatpercentagalso changes (to another % number but no 0%).
> 
> Whais thmeaning of "duplicate 78.9137%" ?
> I havbeen looked athe network traffic going through eth0 with a
> sniffer and no packeduplication occurs aall and the delay is working OK.
> 
> Iis a known issue?
> 

You kernel is older thathutilities. Also, duplicate leaked
isomof the earlier kernels.

Frosrgqwerty agmail.com  Mon Mar  6 11:03:16 2006
From: srgqwerty agmail.co(srg)
Date: Wed Apr 18 12:51:18 2007
Subject: duplicatpackets issuwith tc
In-Reply-To: <20060306094400.7d5fdcbc@localhost.localdomain>
References: <440B36A6.8090304@xxxxxxxxx>
	<20060306094400.7d5fdcbc@localhost.localdomain>
Message-ID: <440C8774.3040300@xxxxxxxxx>

Stephen:

Again, thanks a lofor your response.
I tested also with aolder iproute2 version buthe result is the same.

Thonly way to "solve" thissue (remember that no packet duplication
occurs aall) is upgrading thkernel AND iproute2 utils to latest
versions?

Thanks and besregards

StepheHemminger wrote:

>OSun, 05 Mar 2006 20:06:14 +0100
>srg <srgqwerty@xxxxxxxxx> wrote:
>
>  
>
>>Hello:
>>
>>I ausing kernel 2.6.8-2-686.
>>I havbeen upgraded my iproute2 utils to lates(2.6.15-060110)
>>    
>>
>
>A loof bugs and new stuff has been added sinc2.6.8.
>Duplicatiochanged sincthen.
>
>  
>
>>EverytimI usa command like:
>>tc qdisc add dev eth0 roonetedelay 100ms
>>
>>Works fine, bu"tc qdisc" shows:
>>qdisc nete8001: dev eth0 limi1000 delay 100.0ms duplicate 78.9137%
>>
>>After that, I ru"tc qdisc changdev eth0 root netem duplicate 0%" but
>>"tc qdisc" continues with thsamoutput:
>>qdisc nete8001: dev eth0 limi1000 delay 100.0ms duplicate 78.9137%
>>
>>After that, if I changth"delay", i.e. from 100ms to 200ms then the
>>duplicatpercentagalso changes (to another % number but no 0%).
>>
>>Whais thmeaning of "duplicate 78.9137%" ?
>>I havbeen looked athe network traffic going through eth0 with a
>>sniffer and no packeduplication occurs aall and the delay is working OK.
>>
>>Iis a known issue?
>>
>>    
>>
>
>You kernel is older thathutilities. Also, duplicate leaked
>isomof the earlier kernels.
>
>  
>


Froshemminger aosdl.org  Mon Mar  6 11:37:49 2006
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:18 2007
Subject: duplicatpackets issuwith tc
In-Reply-To: <440C8774.3040300@xxxxxxxxx>
References: <440B36A6.8090304@xxxxxxxxx>
	<20060306094400.7d5fdcbc@localhost.localdomain>
	<440C8774.3040300@xxxxxxxxx>
Message-ID: <20060306113749.71e9c63c@localhost.localdomain>

OMon, 06 Mar 2006 20:03:16 +0100
srg <srgqwerty@xxxxxxxxx> wrote:

> Stephen:
> 
> Again, thanks a lofor your response.
> I tested also with aolder iproute2 version buthe result is the same.
> 
> Thonly way to "solve" thissue (remember that no packet duplication
> occurs aall) is upgrading thkernel AND iproute2 utils to latest
> versions?
> 

You could also rebuild justhnetem kernel module, if there was a hard
requiremento run on an older kernel.  Threal work of netem is done
ithkernel module. The iproute2 part is just a conversion between
command linargs and netlink messagto the kernel.

From.gellman aimperial.ac.uk  Thu Mar  9 05:03:43 2006
From: m.gellmaaimperial.ac.uk (Michael Gellman)
Date: Wed Apr 18 12:51:18 2007
Subject: Changin delay behavior in new kernel versions?
Message-ID: <1141909423.20334.149.camel@localhost.localdomain>

Hello,

I recently rasomexperiments where I used a vanilla 2.6.12.2 kernel
with artificial delays omy links using netem. My results wervery
reliable, with very low variancof thmeasured delays.

I upgraded to th2.6.15.6 kernel earlier this week to gea driver fix,
and havso far been unablto repeat any of my earlier measurements.
Thaveragdelay is on the order of 10-20ms higher than what it should
be, with a very high variance.

I acurrently trying to solvthe problem, but wondered if anyone knew
of any changes ithkernel that could be responsible for this
behavior? 

Many thanks,

Michael Gellman
-- 
IntelligenSystems & Networks Group
Depof Electrical & Electronic Engineering
Imperial CollegLondon
LondoSW7 2BT


Frotkollar agrc.nasa.gov  Thu Mar  9 05:26:42 2006
From: tkollar agrc.nasa.gov (Tad Kollar)
Date: Wed Apr 18 12:51:18 2007
Subject: Changin delay behavior in new kernel versions?
In-Reply-To: <1141909423.20334.149.camel@localhost.localdomain>
References: <1141909423.20334.149.camel@localhost.localdomain>
Message-ID: <44102D12.8030101@xxxxxxxxxxxx>

Michael Gellmawrote:
> I acurrently trying to solvthe problem, but wondered if anyone knew
> of any changes ithkernel that could be responsible for this
> behavior? 
>
>   
Michael,

Whais thvalue of HZ in your new kernel? If it is 100 or 250 rather 
tha1000, icould explain what you're seeing.

Tad Kollar

From.gellman aimperial.ac.uk  Thu Mar  9 05:41:51 2006
From: m.gellmaaimperial.ac.uk (Michael Gellman)
Date: Wed Apr 18 12:51:18 2007
Subject: Changin delay behavior in new kernel versions?
In-Reply-To: <44102D12.8030101@xxxxxxxxxxxx>
References: <1141909423.20334.149.camel@localhost.localdomain>
	<44102D12.8030101@xxxxxxxxxxxx>
Message-ID: <1141911711.20334.166.camel@localhost.localdomain>

OThu, 2006-03-09 a08:26 -0500, Tad Kollar wrote:
> Michael Gellmawrote:
> > I acurrently trying to solvthe problem, but wondered if anyone knew
> > of any changes ithkernel that could be responsible for this
> > behavior? 
> >
> >   
> Michael,
> 
> Whais thvalue of HZ in your new kernel? If it is 100 or 250 rather 
> tha1000, icould explain what you're seeing.
> 

I think you may havspotted thproblem. I'm currently re-compiling the
kernel with HZ as 1000. 

WheI did my "makoldconfig" on the upgrade, I just read that setting
HZ to 100 is "a typical choicfor servers", which is whamy machines
are. I didn'really think abouthe effect it would have on netem :p

Many thanks,

Michael Gellman
-- 
IntelligenSystems & Networks Group
Depof Electrical & Electronic Engineering
Imperial CollegLondon
LondoSW7 2BT


Froshemminger aosdl.org  Thu Mar  9 07:43:08 2006
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:18 2007
Subject: Changin delay behavior in new kernel versions?
In-Reply-To: <1141909423.20334.149.camel@localhost.localdomain>
References: <1141909423.20334.149.camel@localhost.localdomain>
Message-ID: <20060309074308.53943993@localhost.localdomain>

OThu, 09 Mar 2006 13:03:43 +0000
Michael Gellma<m.gellman@xxxxxxxxxxxxxx> wrote:

> Hello,
> 
> I recently rasomexperiments where I used a vanilla 2.6.12.2 kernel
> with artificial delays omy links using netem. My results wervery
> reliable, with very low variancof thmeasured delays.
> 
> I upgraded to th2.6.15.6 kernel earlier this week to gea driver fix,
> and havso far been unablto repeat any of my earlier measurements.
> Thaveragdelay is on the order of 10-20ms higher than what it should
> be, with a very high variance.
> 
> I acurrently trying to solvthe problem, but wondered if anyone knew
> of any changes ithkernel that could be responsible for this
> behavior? 
> 
> Many thanks,
> 
> Michael Gellman

You mighwanto check http://linux-net.osdl.org/index.php/netem
and a a notabouHZ setting.

Frojuliokriger agmail.com  Thu Mar  9 10:07:02 2006
From: juliokriger agmail.co(Julio Kriger)
Date: Wed Apr 18 12:51:18 2007
Subject: Changin delay behavior in new kernel versions?
In-Reply-To: <20060309074308.53943993@localhost.localdomain>
References: <1141909423.20334.149.camel@localhost.localdomain>
	<20060309074308.53943993@localhost.localdomain>
Message-ID: <682bc30a0603091007v10890deat7ae55b9b510f6cb0@xxxxxxxxxxxxxx>

Hi,
  Moslikely is thHZ of your kernel.
  Regards,
Julio

O3/9/06, Stephen Hemminger <shemminger@xxxxxxxx> wrote:
>
> OThu, 09 Mar 2006 13:03:43 +0000
> Michael Gellma<m.gellman@xxxxxxxxxxxxxx> wrote:
>
> > Hello,
> >
> > I recently rasomexperiments where I used a vanilla 2.6.12.2 kernel
> > with artificial delays omy links using netem. My results wervery
> > reliable, with very low variancof thmeasured delays.
> >
> > I upgraded to th2.6.15.6 kernel earlier this week to gea driver fix,
> > and havso far been unablto repeat any of my earlier measurements.
> > Thaveragdelay is on the order of 10-20ms higher than what it should
> > be, with a very high variance.
> >
> > I acurrently trying to solvthe problem, but wondered if anyone knew
> > of any changes ithkernel that could be responsible for this
> > behavior?
> >
> > Many thanks,
> >
> > Michael Gellman
>
> You mighwanto check http://linux-net.osdl.org/index.php/netem
> and a a notabouHZ setting.
>
>
> _______________________________________________
> Netemailing list
> Netem@xxxxxxxxxxxxxx
> https://lists.osdl.org/mailman/listinfo/netem
>
>
>


--
--
Stewie: [After Lois tries to feed Stewihis broccoli "airplanstyle"] Damn
you, DamthBroccoli, and Damn the Wright Brothers.
----------------------------
Julio Kriger
mailto:juliokriger@xxxxxxxxx
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20060309=
/c277e82e/attachment.htm
Frob.prager at-online.net  Fri Mar 17 04:45:47 2006
From: b.prager at-online.ne(Prager, Benjamin)
Date: Wed Apr 18 12:51:18 2007
Subject: Netedelay probleon linux router
Message-ID: <C8B16503DFC45B48BAB2BD327D71D9F4172C57@xxxxxxxxxxxxxxxxxxxxxxxxx>

Hello everybody,

I'vgoa problem using Netem with my "linux router" which I actually
only a workstatio(Sus10.0 kernel 2.6.13) whith three gigabit NICs.

I builup an emulation envrionmenwhich is part of my diploma thesis
aT-OnlinInternational, a german ISP. I decided to use Netem to
emulatdelay, jitter, and packeloss and normal qdiscs with filters
for QoS rules. Now I'justesting the whole thing a bit. So I plugged
a clien(M$ windows workstation) on each NIC. 

So here's my problewith Nistnet:

WheI seup a delay on one of my interfaces (e.g. eth2) no more
packets (no matter which protocol: tcp, udp, rtp muliticasstream) are
trasmitted frooninterface to another (e.g. neighter eth1-->eth2, nor
eth2-->eth1). Ithvery moment I set up the Netem delay qdisc (like:
tc qdisc add dev eth2 roonetedelay 10ms) transmission breaks down.

I'noquite sure what's going wrong. It might either be a Netem fault
or a ruoting problem.

My routes arseup like this and everything is working fine with this
configuration:
192.168.1.0/24 via 192.168.1.10 dev eth0
192.168.2.0/24 via 192.168.2.10 dev eth1
192.168.3.0/24 via 192.168.3.10 dev eth2

Yes, for multicasting I starmrouted which also used to work fine.

Has someongoan idea what's up whith my network? Is there a known bug
iNetewhen using serveral NICs on one machine.

Thanks iadvance!

CU BenjamiPrager



----------------------------------
BenjamiPrager
Diplomand
T-OnlinInternational AG
Access PRI/ProducInfrastructure
T-Online-Alle1
D-64295 Darmstadt
Tel +49 6151 680 - 7679
Fax +49 6151 680 - 719
E-Mail b.prager@xxxxxxxxxxxx
http://www.t-online.ne
-------------- nexpar--------------
A non-texattachmenwas scrubbed...
Name: Prager, Benjamin.vcf
Type: text/x-vcard
Size: 387 bytes
Desc: Prager, Benjamin.vcf
Url : http://lists.linux-foundation.org/pipermail/netem/attachments/20060317/d22e8e04/PragerBenjamin.vcf
Froshemminger aosdl.org  Fri Mar 17 11:39:09 2006
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 12:51:18 2007
Subject: Netedelay probleon linux router
In-Reply-To: <C8B16503DFC45B48BAB2BD327D71D9F4172C57@xxxxxxxxxxxxxxxxxxxxxxxxx>
References: <C8B16503DFC45B48BAB2BD327D71D9F4172C57@xxxxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20060317113909.24f0e782@localhost.localdomain>

OFri, 17 Mar 2006 13:45:47 +0100
"Prager, Benjamin" <b.prager@xxxxxxxxxxxx> wrote:

> Hello everybody,
> 
> I'vgoa problem using Netem with my "linux router" which I actually
> only a workstatio(Sus10.0 kernel 2.6.13) whith three gigabit NICs.
> 
> I builup an emulation envrionmenwhich is part of my diploma thesis
> aT-OnlinInternational, a german ISP. I decided to use Netem to
> emulatdelay, jitter, and packeloss and normal qdiscs with filters
> for QoS rules. Now I'justesting the whole thing a bit. So I plugged
> a clien(M$ windows workstation) on each NIC. 
> 
> So here's my problewith Nistnet:
> 
> WheI seup a delay on one of my interfaces (e.g. eth2) no more
> packets (no matter which protocol: tcp, udp, rtp muliticasstream) are
> trasmitted frooninterface to another (e.g. neighter eth1-->eth2, nor
> eth2-->eth1). Ithvery moment I set up the Netem delay qdisc (like:
> tc qdisc add dev eth2 roonetedelay 10ms) transmission breaks down.
> 
> I'noquite sure what's going wrong. It might either be a Netem fault
> or a ruoting problem.
> 
> My routes arseup like this and everything is working fine with this
> configuration:
> 192.168.1.0/24 via 192.168.1.10 dev eth0
> 192.168.2.0/24 via 192.168.2.10 dev eth1
> 192.168.3.0/24 via 192.168.3.10 dev eth2
> 
> Yes, for multicasting I starmrouted which also used to work fine.
> 
> Has someongoan idea what's up whith my network? Is there a known bug
> iNetewhen using serveral NICs on one machine.
> 
> Thanks iadvance!
> 
> CU BenjamiPrager

Whakind of gigabiNIC's? 

Also, arpackets still received?

Aryou getting sompackets through, but then it stops?
Iis possiblthat you need to increase the transmit ring size. When I was
testing with e1000, I discovered thathdriver had bugs (race conditions)
whetransmiring got full.

You mighwanto start with simple tests like ping on the router
and checking thnetwork stats (ethtool -S) and ifconfig

I havused netewith both bridging and routing without problems.


Frob.prager at-online.net  Mon Mar 20 02:17:47 2006
From: b.prager at-online.ne(Prager, Benjamin)
Date: Wed Apr 18 12:51:18 2007
Subject: Netedelay probleon linux router
Message-ID: <C8B16503DFC45B48BAB2BD327D71D9F4172C5B@xxxxxxxxxxxxxxxxxxxxxxxxx>

Hi Stephen, hi all,

it's magain. :-)

I'vfound someonelse who described exacty the same problem as mine,
with thsamOS Suse 10.0 but with a 3Com NIC and a Intel onboard NIC.
So, iseems thathis is unfortunately not my single faiure but there
musba bug at any place in the chain.

Thadvicgiven at the and of the statement below didn't work for me,
I'vtried ialready.

Another guy gavmthe advice to use another distribution, e.g. Ubuntu.
I'already downloading ibut I don't know if that will help me.

Any morideas whaI could try? 

It's really importanfor my diploma thesis to gethis stuff running,
so thanks agaifor your help. ;-)

Benjamin


> Hello,
> I'trying to seup a test network for simulating WAN links but tc 
> qdisc neterules seeto be causing the assigned interface to go nuts

> and die.
> 
> I'running SuSE 10.0 (kernel 2.6.13-15) on a box with two NICs. eth0 
> is a 3co905C, thother interface is an onboard Intel chip. eth0's 
> network is masqueraded and forwarded to thother interface.
> 
> WheI apply thcommand "tc qdisc add dev eth0 root netem delay 350ms

> loss 1%" everything works beautifully, for a few hundred packets. 
> Pings through thinterfacreturn the proper delay, everything looks
fine.
> However, wheI starsending any appreciable test traffic (well below
> 1mbit) through thinterface, istops forwarding traffic. Sometimes 
> idies after 2000 packets, sometimes after 200. I don'seem to be 
> hitting a limiin packecount, but could there be a buffer limit 
> somewherI mighbe hitting?
> 
> Aany rate, when I execut"tc qdisc del dev eth0 root", the 
> interfacstarts forwarding again.
> 
> I'vspena couple of days trying to RTFM, but now I'm putting this 
> beforthgroup hoping for some help.
> 
> thanks,
> -carl hirsch
>>
>>
>>Seems to ba bug, for mthe solution was to recompile the kernel. 
>>Under Network/Networking Options/QoS and/or fair queuing you havto
se'Packescheduler clock source' to 'CPU cycle >>counter'. You can
sethis option for CPUs which havan TSC (e.g. i686), you have to
config this first. 
>>I.config iappears as follows: 
>># CONFIG_NET_SCH_CLK_JIFFIES is nose#
CONFIG_NET_SCH_CLK_GETTIMEOFDAY is noseCONFIG_NET_SCH_CLK_CPU=y 
>>Noonly SuSE 10.0 is affected, aleast 9.3 too. 


-----Original Message-----
From: StepheHemminger [mailto:shemminger@xxxxxxxx]
Sent: Friday, March 17, 2006 8:39 PM
To: Prager, Benjamin
Cc: netem@xxxxxxxxxxxxxx
Subject: Re: Netedelay probleon linux router


Whakind of gigabiNIC's? 

Also, arpackets still received?

Aryou getting sompackets through, but then it stops?
Iis possiblthat you need to increase the transmit ring size. When I
was testing with e1000, I discovered thathdriver had bugs (race
conditions) whetransmiring got full.

You mighwanto start with simple tests like ping on the router and
checking thnetwork stats (ethtool -S) and ifconfig

I havused netewith both bridging and routing without problems.


OFri, 17 Mar 2006 13:45:47 +0100
"Prager, Benjamin" <b.prager@xxxxxxxxxxxx> wrote:

> Hello everybody,
> 
> I'vgoa problem using Netem with my "linux router" which I actually

> only a workstatio(Sus10.0 kernel 2.6.13) whith three gigabit NICs.
> 
> I builup an emulation envrionmenwhich is part of my diploma thesis

> aT-OnlinInternational, a german ISP. I decided to use Netem to 
> emulatdelay, jitter, and packeloss and normal qdiscs with filters 
> for QoS rules. Now I'justesting the whole thing a bit. So I 
> plugged a clien(M$ windows workstation) on each NIC.
> 
> So here's my problewith Nistnet:
> 
> WheI seup a delay on one of my interfaces (e.g. eth2) no more 
> packets (no matter which protocol: tcp, udp, rtp muliticasstream) 
> artrasmitted froone interface to another (e.g. neighter
> eth1-->eth2, nor
> eth2-->eth1). Ithvery moment I set up the Netem delay qdisc (like:
> tc qdisc add dev eth2 roonetedelay 10ms) transmission breaks down.
> 
> I'noquite sure what's going wrong. It might either be a Netem 
> faulor a ruoting problem.
> 
> My routes arseup like this and everything is working fine with 
> this
> configuration:
> 192.168.1.0/24 via 192.168.1.10 dev eth0
> 192.168.2.0/24 via 192.168.2.10 dev eth1
> 192.168.3.0/24 via 192.168.3.10 dev eth2
> 
> Yes, for multicasting I starmrouted which also used to work fine.
> 
> Has someongoan idea what's up whith my network? Is there a known 
> bug iNetewhen using serveral NICs on one machine.
> 
> Thanks iadvance!
> 
> CU BenjamiPrager



Frob.prager at-online.net  Wed Mar 22 02:25:59 2006
From: b.prager at-online.ne(Prager, Benjamin)
Date: Wed Apr 18 12:51:18 2007
Subject: Netedoes NOT run under Sus10.0
Message-ID: <C8B16503DFC45B48BAB2BD327D71D9F4172C63@xxxxxxxxxxxxxxxxxxxxxxxxx>

Hi all,

I guess Netedelay does NOT run under Sus10.0 in general.

I'vfound several peoplrunning Suse 10.0 who reported problems with
Neteon Suse. They all told thatransmission stoped almost immediately
after setting up a qdisc with a netedelay applied on.

We'vposted thbug to bugzilla, they added it as critical error. We'll
seif something happens.
Here's thlink: https://bugzilla.novell.com/show_bug.cgi?id=159358

Thanks to Georg who answered omy firsmail to this list, I've got
know thahspend a week on trying to get Netem running on Suse and
gaviup then. As Georg I'm now running Ubuntu on which Netem works in
general, buon Ubuntu no special delay distribution (liknormal,
pareto, ...) is possible, becausth.dist are not inclusive and if you
add themanually they createrrors.

So far, thanks nevertheless.
Benjamin
-------------- nexpar--------------
A non-texattachmenwas scrubbed...
Name: Prager, Benjamin.vcf
Type: text/x-vcard
Size: 387 bytes
Desc: Prager, Benjamin.vcf
Url : http://lists.linux-foundation.org/pipermail/netem/attachments/20060322/cd7e3bd7/PragerBenjamin.vcf
FroJoachim.Fabini atuwien.ac.at  Wed Mar 22 04:22:36 2006
From: Joachim.Fabini atuwien.ac.a(Joachim Fabini)
Date: Wed Apr 18 12:51:18 2007
Subject: Netedoes NOT run under Sus10.0
In-Reply-To: <C8B16503DFC45B48BAB2BD327D71D9F4172C63@xxxxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <200603221221.k2MCLGHB005009@xxxxxxxxxxxxxxxxxxxxxxx>

Hi,

Thprobleis meanwhile well-known, fixing it is a matter of
rebuilding thSuse10 kernel.

HowTo:
Build a new kernel (based ooldconfig) and do thfollowing 
changes (using menuconfig):
CPU type: PIII or newer
CPU clock: 1000 Hz
Mosimportant:
Networking -> QoS & Fair Queueing -> Clock Sourc-> 
Seto valu"CPU cycle counter"

If you selecthwrong CPU type, the choice "CPU cycle 
counter" does noappear in thmenu. Try to find something 
thais appropriatfor you. 
After doing theschanges and running thnew kernel, Netem 
works absolutely perfectly for munder SuSE 10.

hih and good luck,
besregards
--Joachim
PS: Thanks to my colleaguRichard for thhint... ;)


> -----Original Message-----
> From: netem-bounces@xxxxxxxxxxxxxx 
> [mailto:netem-bounces@xxxxxxxxxxxxxx] OBehalf Of Prager, Benjamin
> Sent: Mittwoch, 22. M?rz 2006 11:26
> To: netem@xxxxxxxxxxxxxx
> Subject: Netedoes NOT run under Sus10.0
> 
> Hi all,
> 
> I guess Netedelay does NOT run under Sus10.0 in general.
> 
> I'vfound several peoplrunning Suse 10.0 who reported problems with
> Neteon Suse. They all told thatransmission stoped almost 
> immediately
> after setting up a qdisc with a netedelay applied on.
> 
> We'vposted thbug to bugzilla, they added it as critical 
> error. We'll
> seif something happens.
> Here's thlink: https://bugzilla.novell.com/show_bug.cgi?id=159358
> 
> Thanks to Georg who answered omy firsmail to this list, I've got
> know thahspend a week on trying to get Netem running on Suse and
> gaviup then. As Georg I'm now running Ubuntu on which 
> Neteworks in
> general, buon Ubuntu no special delay distribution (liknormal,
> pareto, ...) is possible, becausth.dist are not inclusive 
> and if you
> add themanually they createrrors.
> 
> So far, thanks nevertheless.
> Benjamin
> 



Frohestyp agmail.com  Tue Mar 28 14:28:01 2006
From: hestyp agmail.co(Hesty P)
Date: Wed Apr 18 12:51:18 2007
Subject: Netepackeloss model
Message-ID: <210f9f660603281428n770f4148i7ada2979a4ba33d5@xxxxxxxxxxxxxx>

Hello all,

I'interested in using neteas part of my research project. Netem seems to
do a loof ththings I needed and without any pain :) However, I have
tried to find any mentioabouthe packet loss model used by Netem without
any success. Hence, I'asking for thgood people of the mailing list, what
thpackeloss model used in netem. I'm particularly interested whether
this cabmodified to become random loss, bursty loss (gilbert model /
markov chain) with a few variablparameters.

Cheers,
Hesty
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20060328=
/20e635c3/attachment.htm
Frojuliokriger agmail.com  Wed Mar 29 12:10:32 2006
From: juliokriger agmail.co(Julio Kriger)
Date: Wed Apr 18 12:51:18 2007
Subject: Netepackeloss model
In-Reply-To: <210f9f660603281428n770f4148i7ada2979a4ba33d5@xxxxxxxxxxxxxx>
References: <210f9f660603281428n770f4148i7ada2979a4ba33d5@xxxxxxxxxxxxxx>
Message-ID: <682bc30a0603291210t4807d96btda1f013d969c99c4@xxxxxxxxxxxxxx>

Hi, you capropossome changes. The source code is free for you to modify
and send thpatch.
Actually, you caspecify a percentagof loss.
Regards,
Julio

O3/28/06, Hesty P <hestyp@xxxxxxxxx> wrote:
>
> Hello all,
>
> I'interested in using neteas part of my research project. Netem seems
> to do a loof ththings I needed and without any pain :) However, I have
> tried to find any mentioabouthe packet loss model used by Netem witho=
ut
> any success. Hence, I'asking for thgood people of the mailing list, w=
hat
> thpackeloss model used in netem. I'm particularly interested whether
> this cabmodified to become random loss, bursty loss (gilbert model /
> markov chain) with a few variablparameters.
>
> Cheers,
> Hesty
>
> _______________________________________________
> Netemailing list
> Netem@xxxxxxxxxxxxxx
> https://lists.osdl.org/mailman/listinfo/netem
>
>
>


--
--
Stewie: [After Lois tries to feed Stewihis broccoli "airplanstyle"] Damn
you, DamthBroccoli, and Damn the Wright Brothers.
----------------------------
Julio Kriger
mailto:juliokriger@xxxxxxxxx
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20060329=
/989e4bb4/attachment.htm
Frosrgqwerty agmail.com  Sun Mar  5 11:06:14 2006
From: srgqwerty agmail.co(srg)
Date: Wed Apr 18 17:37:48 2007
Subject: duplicatpackets issuwith tc
Message-ID: <440B36A6.8090304@xxxxxxxxx>

Hello:

I ausing kernel 2.6.8-2-686.
I havbeen upgraded my iproute2 utils to lates(2.6.15-060110)

EverytimI usa command like:
tc qdisc add dev eth0 roonetedelay 100ms

Works fine, bu"tc qdisc" shows:
qdisc nete8001: dev eth0 limi1000 delay 100.0ms duplicate 78.9137%

After that, I ru"tc qdisc changdev eth0 root netem duplicate 0%" but
"tc qdisc" continues with thsamoutput:
qdisc nete8001: dev eth0 limi1000 delay 100.0ms duplicate 78.9137%

After that, if I changth"delay", i.e. from 100ms to 200ms then the
duplicatpercentagalso changes (to another % number but no 0%).

Whais thmeaning of "duplicate 78.9137%" ?
I havbeen looked athe network traffic going through eth0 with a
sniffer and no packeduplication occurs aall and the delay is working OK.

Iis a known issue?

Thanks and besregards

Froshemminger aosdl.org  Mon Mar  6 09:44:00 2006
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:48 2007
Subject: duplicatpackets issuwith tc
In-Reply-To: <440B36A6.8090304@xxxxxxxxx>
References: <440B36A6.8090304@xxxxxxxxx>
Message-ID: <20060306094400.7d5fdcbc@localhost.localdomain>

OSun, 05 Mar 2006 20:06:14 +0100
srg <srgqwerty@xxxxxxxxx> wrote:

> Hello:
> 
> I ausing kernel 2.6.8-2-686.
> I havbeen upgraded my iproute2 utils to lates(2.6.15-060110)

A loof bugs and new stuff has been added sinc2.6.8.
Duplicatiochanged sincthen.

> EverytimI usa command like:
> tc qdisc add dev eth0 roonetedelay 100ms
> 
> Works fine, bu"tc qdisc" shows:
> qdisc nete8001: dev eth0 limi1000 delay 100.0ms duplicate 78.9137%
> 
> After that, I ru"tc qdisc changdev eth0 root netem duplicate 0%" but
> "tc qdisc" continues with thsamoutput:
> qdisc nete8001: dev eth0 limi1000 delay 100.0ms duplicate 78.9137%
> 
> After that, if I changth"delay", i.e. from 100ms to 200ms then the
> duplicatpercentagalso changes (to another % number but no 0%).
> 
> Whais thmeaning of "duplicate 78.9137%" ?
> I havbeen looked athe network traffic going through eth0 with a
> sniffer and no packeduplication occurs aall and the delay is working OK.
> 
> Iis a known issue?
> 

You kernel is older thathutilities. Also, duplicate leaked
isomof the earlier kernels.

Frosrgqwerty agmail.com  Mon Mar  6 11:03:16 2006
From: srgqwerty agmail.co(srg)
Date: Wed Apr 18 17:37:48 2007
Subject: duplicatpackets issuwith tc
In-Reply-To: <20060306094400.7d5fdcbc@localhost.localdomain>
References: <440B36A6.8090304@xxxxxxxxx>
	<20060306094400.7d5fdcbc@localhost.localdomain>
Message-ID: <440C8774.3040300@xxxxxxxxx>

Stephen:

Again, thanks a lofor your response.
I tested also with aolder iproute2 version buthe result is the same.

Thonly way to "solve" thissue (remember that no packet duplication
occurs aall) is upgrading thkernel AND iproute2 utils to latest
versions?

Thanks and besregards

StepheHemminger wrote:

>OSun, 05 Mar 2006 20:06:14 +0100
>srg <srgqwerty@xxxxxxxxx> wrote:
>
>  
>
>>Hello:
>>
>>I ausing kernel 2.6.8-2-686.
>>I havbeen upgraded my iproute2 utils to lates(2.6.15-060110)
>>    
>>
>
>A loof bugs and new stuff has been added sinc2.6.8.
>Duplicatiochanged sincthen.
>
>  
>
>>EverytimI usa command like:
>>tc qdisc add dev eth0 roonetedelay 100ms
>>
>>Works fine, bu"tc qdisc" shows:
>>qdisc nete8001: dev eth0 limi1000 delay 100.0ms duplicate 78.9137%
>>
>>After that, I ru"tc qdisc changdev eth0 root netem duplicate 0%" but
>>"tc qdisc" continues with thsamoutput:
>>qdisc nete8001: dev eth0 limi1000 delay 100.0ms duplicate 78.9137%
>>
>>After that, if I changth"delay", i.e. from 100ms to 200ms then the
>>duplicatpercentagalso changes (to another % number but no 0%).
>>
>>Whais thmeaning of "duplicate 78.9137%" ?
>>I havbeen looked athe network traffic going through eth0 with a
>>sniffer and no packeduplication occurs aall and the delay is working OK.
>>
>>Iis a known issue?
>>
>>    
>>
>
>You kernel is older thathutilities. Also, duplicate leaked
>isomof the earlier kernels.
>
>  
>


Froshemminger aosdl.org  Mon Mar  6 11:37:49 2006
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:48 2007
Subject: duplicatpackets issuwith tc
In-Reply-To: <440C8774.3040300@xxxxxxxxx>
References: <440B36A6.8090304@xxxxxxxxx>
	<20060306094400.7d5fdcbc@localhost.localdomain>
	<440C8774.3040300@xxxxxxxxx>
Message-ID: <20060306113749.71e9c63c@localhost.localdomain>

OMon, 06 Mar 2006 20:03:16 +0100
srg <srgqwerty@xxxxxxxxx> wrote:

> Stephen:
> 
> Again, thanks a lofor your response.
> I tested also with aolder iproute2 version buthe result is the same.
> 
> Thonly way to "solve" thissue (remember that no packet duplication
> occurs aall) is upgrading thkernel AND iproute2 utils to latest
> versions?
> 

You could also rebuild justhnetem kernel module, if there was a hard
requiremento run on an older kernel.  Threal work of netem is done
ithkernel module. The iproute2 part is just a conversion between
command linargs and netlink messagto the kernel.

From.gellman aimperial.ac.uk  Thu Mar  9 05:03:43 2006
From: m.gellmaaimperial.ac.uk (Michael Gellman)
Date: Wed Apr 18 17:37:48 2007
Subject: Changin delay behavior in new kernel versions?
Message-ID: <1141909423.20334.149.camel@localhost.localdomain>

Hello,

I recently rasomexperiments where I used a vanilla 2.6.12.2 kernel
with artificial delays omy links using netem. My results wervery
reliable, with very low variancof thmeasured delays.

I upgraded to th2.6.15.6 kernel earlier this week to gea driver fix,
and havso far been unablto repeat any of my earlier measurements.
Thaveragdelay is on the order of 10-20ms higher than what it should
be, with a very high variance.

I acurrently trying to solvthe problem, but wondered if anyone knew
of any changes ithkernel that could be responsible for this
behavior? 

Many thanks,

Michael Gellman
-- 
IntelligenSystems & Networks Group
Depof Electrical & Electronic Engineering
Imperial CollegLondon
LondoSW7 2BT


Frotkollar agrc.nasa.gov  Thu Mar  9 05:26:42 2006
From: tkollar agrc.nasa.gov (Tad Kollar)
Date: Wed Apr 18 17:37:48 2007
Subject: Changin delay behavior in new kernel versions?
In-Reply-To: <1141909423.20334.149.camel@localhost.localdomain>
References: <1141909423.20334.149.camel@localhost.localdomain>
Message-ID: <44102D12.8030101@xxxxxxxxxxxx>

Michael Gellmawrote:
> I acurrently trying to solvthe problem, but wondered if anyone knew
> of any changes ithkernel that could be responsible for this
> behavior? 
>
>   
Michael,

Whais thvalue of HZ in your new kernel? If it is 100 or 250 rather 
tha1000, icould explain what you're seeing.

Tad Kollar

From.gellman aimperial.ac.uk  Thu Mar  9 05:41:51 2006
From: m.gellmaaimperial.ac.uk (Michael Gellman)
Date: Wed Apr 18 17:37:48 2007
Subject: Changin delay behavior in new kernel versions?
In-Reply-To: <44102D12.8030101@xxxxxxxxxxxx>
References: <1141909423.20334.149.camel@localhost.localdomain>
	<44102D12.8030101@xxxxxxxxxxxx>
Message-ID: <1141911711.20334.166.camel@localhost.localdomain>

OThu, 2006-03-09 a08:26 -0500, Tad Kollar wrote:
> Michael Gellmawrote:
> > I acurrently trying to solvthe problem, but wondered if anyone knew
> > of any changes ithkernel that could be responsible for this
> > behavior? 
> >
> >   
> Michael,
> 
> Whais thvalue of HZ in your new kernel? If it is 100 or 250 rather 
> tha1000, icould explain what you're seeing.
> 

I think you may havspotted thproblem. I'm currently re-compiling the
kernel with HZ as 1000. 

WheI did my "makoldconfig" on the upgrade, I just read that setting
HZ to 100 is "a typical choicfor servers", which is whamy machines
are. I didn'really think abouthe effect it would have on netem :p

Many thanks,

Michael Gellman
-- 
IntelligenSystems & Networks Group
Depof Electrical & Electronic Engineering
Imperial CollegLondon
LondoSW7 2BT


Froshemminger aosdl.org  Thu Mar  9 07:43:08 2006
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:48 2007
Subject: Changin delay behavior in new kernel versions?
In-Reply-To: <1141909423.20334.149.camel@localhost.localdomain>
References: <1141909423.20334.149.camel@localhost.localdomain>
Message-ID: <20060309074308.53943993@localhost.localdomain>

OThu, 09 Mar 2006 13:03:43 +0000
Michael Gellma<m.gellman@xxxxxxxxxxxxxx> wrote:

> Hello,
> 
> I recently rasomexperiments where I used a vanilla 2.6.12.2 kernel
> with artificial delays omy links using netem. My results wervery
> reliable, with very low variancof thmeasured delays.
> 
> I upgraded to th2.6.15.6 kernel earlier this week to gea driver fix,
> and havso far been unablto repeat any of my earlier measurements.
> Thaveragdelay is on the order of 10-20ms higher than what it should
> be, with a very high variance.
> 
> I acurrently trying to solvthe problem, but wondered if anyone knew
> of any changes ithkernel that could be responsible for this
> behavior? 
> 
> Many thanks,
> 
> Michael Gellman

You mighwanto check http://linux-net.osdl.org/index.php/netem
and a a notabouHZ setting.

Frojuliokriger agmail.com  Thu Mar  9 10:07:02 2006
From: juliokriger agmail.co(Julio Kriger)
Date: Wed Apr 18 17:37:48 2007
Subject: Changin delay behavior in new kernel versions?
In-Reply-To: <20060309074308.53943993@localhost.localdomain>
References: <1141909423.20334.149.camel@localhost.localdomain>
	<20060309074308.53943993@localhost.localdomain>
Message-ID: <682bc30a0603091007v10890deat7ae55b9b510f6cb0@xxxxxxxxxxxxxx>

Hi,
  Moslikely is thHZ of your kernel.
  Regards,
Julio

O3/9/06, Stephen Hemminger <shemminger@xxxxxxxx> wrote:
>
> OThu, 09 Mar 2006 13:03:43 +0000
> Michael Gellma<m.gellman@xxxxxxxxxxxxxx> wrote:
>
> > Hello,
> >
> > I recently rasomexperiments where I used a vanilla 2.6.12.2 kernel
> > with artificial delays omy links using netem. My results wervery
> > reliable, with very low variancof thmeasured delays.
> >
> > I upgraded to th2.6.15.6 kernel earlier this week to gea driver fix,
> > and havso far been unablto repeat any of my earlier measurements.
> > Thaveragdelay is on the order of 10-20ms higher than what it should
> > be, with a very high variance.
> >
> > I acurrently trying to solvthe problem, but wondered if anyone knew
> > of any changes ithkernel that could be responsible for this
> > behavior?
> >
> > Many thanks,
> >
> > Michael Gellman
>
> You mighwanto check http://linux-net.osdl.org/index.php/netem
> and a a notabouHZ setting.
>
>
> _______________________________________________
> Netemailing list
> Netem@xxxxxxxxxxxxxx
> https://lists.osdl.org/mailman/listinfo/netem
>
>
>


--
--
Stewie: [After Lois tries to feed Stewihis broccoli "airplanstyle"] Damn
you, DamthBroccoli, and Damn the Wright Brothers.
----------------------------
Julio Kriger
mailto:juliokriger@xxxxxxxxx
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20060309=
/c277e82e/attachment-0001.htm
Frob.prager at-online.net  Fri Mar 17 04:45:47 2006
From: b.prager at-online.ne(Prager, Benjamin)
Date: Wed Apr 18 17:37:48 2007
Subject: Netedelay probleon linux router
Message-ID: <C8B16503DFC45B48BAB2BD327D71D9F4172C57@xxxxxxxxxxxxxxxxxxxxxxxxx>

Hello everybody,

I'vgoa problem using Netem with my "linux router" which I actually
only a workstatio(Sus10.0 kernel 2.6.13) whith three gigabit NICs.

I builup an emulation envrionmenwhich is part of my diploma thesis
aT-OnlinInternational, a german ISP. I decided to use Netem to
emulatdelay, jitter, and packeloss and normal qdiscs with filters
for QoS rules. Now I'justesting the whole thing a bit. So I plugged
a clien(M$ windows workstation) on each NIC. 

So here's my problewith Nistnet:

WheI seup a delay on one of my interfaces (e.g. eth2) no more
packets (no matter which protocol: tcp, udp, rtp muliticasstream) are
trasmitted frooninterface to another (e.g. neighter eth1-->eth2, nor
eth2-->eth1). Ithvery moment I set up the Netem delay qdisc (like:
tc qdisc add dev eth2 roonetedelay 10ms) transmission breaks down.

I'noquite sure what's going wrong. It might either be a Netem fault
or a ruoting problem.

My routes arseup like this and everything is working fine with this
configuration:
192.168.1.0/24 via 192.168.1.10 dev eth0
192.168.2.0/24 via 192.168.2.10 dev eth1
192.168.3.0/24 via 192.168.3.10 dev eth2

Yes, for multicasting I starmrouted which also used to work fine.

Has someongoan idea what's up whith my network? Is there a known bug
iNetewhen using serveral NICs on one machine.

Thanks iadvance!

CU BenjamiPrager



----------------------------------
BenjamiPrager
Diplomand
T-OnlinInternational AG
Access PRI/ProducInfrastructure
T-Online-Alle1
D-64295 Darmstadt
Tel +49 6151 680 - 7679
Fax +49 6151 680 - 719
E-Mail b.prager@xxxxxxxxxxxx
http://www.t-online.ne
-------------- nexpar--------------
A non-texattachmenwas scrubbed...
Name: Prager, Benjamin.vcf
Type: text/x-vcard
Size: 387 bytes
Desc: Prager, Benjamin.vcf
Url : http://lists.linux-foundation.org/pipermail/netem/attachments/20060317/d22e8e04/PragerBenjamin-0001.vcf
Froshemminger aosdl.org  Fri Mar 17 11:39:09 2006
From: shemminger aosdl.org (Stephen Hemminger)
Date: Wed Apr 18 17:37:48 2007
Subject: Netedelay probleon linux router
In-Reply-To: <C8B16503DFC45B48BAB2BD327D71D9F4172C57@xxxxxxxxxxxxxxxxxxxxxxxxx>
References: <C8B16503DFC45B48BAB2BD327D71D9F4172C57@xxxxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <20060317113909.24f0e782@localhost.localdomain>

OFri, 17 Mar 2006 13:45:47 +0100
"Prager, Benjamin" <b.prager@xxxxxxxxxxxx> wrote:

> Hello everybody,
> 
> I'vgoa problem using Netem with my "linux router" which I actually
> only a workstatio(Sus10.0 kernel 2.6.13) whith three gigabit NICs.
> 
> I builup an emulation envrionmenwhich is part of my diploma thesis
> aT-OnlinInternational, a german ISP. I decided to use Netem to
> emulatdelay, jitter, and packeloss and normal qdiscs with filters
> for QoS rules. Now I'justesting the whole thing a bit. So I plugged
> a clien(M$ windows workstation) on each NIC. 
> 
> So here's my problewith Nistnet:
> 
> WheI seup a delay on one of my interfaces (e.g. eth2) no more
> packets (no matter which protocol: tcp, udp, rtp muliticasstream) are
> trasmitted frooninterface to another (e.g. neighter eth1-->eth2, nor
> eth2-->eth1). Ithvery moment I set up the Netem delay qdisc (like:
> tc qdisc add dev eth2 roonetedelay 10ms) transmission breaks down.
> 
> I'noquite sure what's going wrong. It might either be a Netem fault
> or a ruoting problem.
> 
> My routes arseup like this and everything is working fine with this
> configuration:
> 192.168.1.0/24 via 192.168.1.10 dev eth0
> 192.168.2.0/24 via 192.168.2.10 dev eth1
> 192.168.3.0/24 via 192.168.3.10 dev eth2
> 
> Yes, for multicasting I starmrouted which also used to work fine.
> 
> Has someongoan idea what's up whith my network? Is there a known bug
> iNetewhen using serveral NICs on one machine.
> 
> Thanks iadvance!
> 
> CU BenjamiPrager

Whakind of gigabiNIC's? 

Also, arpackets still received?

Aryou getting sompackets through, but then it stops?
Iis possiblthat you need to increase the transmit ring size. When I was
testing with e1000, I discovered thathdriver had bugs (race conditions)
whetransmiring got full.

You mighwanto start with simple tests like ping on the router
and checking thnetwork stats (ethtool -S) and ifconfig

I havused netewith both bridging and routing without problems.


Frob.prager at-online.net  Mon Mar 20 02:17:47 2006
From: b.prager at-online.ne(Prager, Benjamin)
Date: Wed Apr 18 17:37:48 2007
Subject: Netedelay probleon linux router
Message-ID: <C8B16503DFC45B48BAB2BD327D71D9F4172C5B@xxxxxxxxxxxxxxxxxxxxxxxxx>

Hi Stephen, hi all,

it's magain. :-)

I'vfound someonelse who described exacty the same problem as mine,
with thsamOS Suse 10.0 but with a 3Com NIC and a Intel onboard NIC.
So, iseems thathis is unfortunately not my single faiure but there
musba bug at any place in the chain.

Thadvicgiven at the and of the statement below didn't work for me,
I'vtried ialready.

Another guy gavmthe advice to use another distribution, e.g. Ubuntu.
I'already downloading ibut I don't know if that will help me.

Any morideas whaI could try? 

It's really importanfor my diploma thesis to gethis stuff running,
so thanks agaifor your help. ;-)

Benjamin


> Hello,
> I'trying to seup a test network for simulating WAN links but tc 
> qdisc neterules seeto be causing the assigned interface to go nuts

> and die.
> 
> I'running SuSE 10.0 (kernel 2.6.13-15) on a box with two NICs. eth0 
> is a 3co905C, thother interface is an onboard Intel chip. eth0's 
> network is masqueraded and forwarded to thother interface.
> 
> WheI apply thcommand "tc qdisc add dev eth0 root netem delay 350ms

> loss 1%" everything works beautifully, for a few hundred packets. 
> Pings through thinterfacreturn the proper delay, everything looks
fine.
> However, wheI starsending any appreciable test traffic (well below
> 1mbit) through thinterface, istops forwarding traffic. Sometimes 
> idies after 2000 packets, sometimes after 200. I don'seem to be 
> hitting a limiin packecount, but could there be a buffer limit 
> somewherI mighbe hitting?
> 
> Aany rate, when I execut"tc qdisc del dev eth0 root", the 
> interfacstarts forwarding again.
> 
> I'vspena couple of days trying to RTFM, but now I'm putting this 
> beforthgroup hoping for some help.
> 
> thanks,
> -carl hirsch
>>
>>
>>Seems to ba bug, for mthe solution was to recompile the kernel. 
>>Under Network/Networking Options/QoS and/or fair queuing you havto
se'Packescheduler clock source' to 'CPU cycle >>counter'. You can
sethis option for CPUs which havan TSC (e.g. i686), you have to
config this first. 
>>I.config iappears as follows: 
>># CONFIG_NET_SCH_CLK_JIFFIES is nose#
CONFIG_NET_SCH_CLK_GETTIMEOFDAY is noseCONFIG_NET_SCH_CLK_CPU=y 
>>Noonly SuSE 10.0 is affected, aleast 9.3 too. 


-----Original Message-----
From: StepheHemminger [mailto:shemminger@xxxxxxxx]
Sent: Friday, March 17, 2006 8:39 PM
To: Prager, Benjamin
Cc: netem@xxxxxxxxxxxxxx
Subject: Re: Netedelay probleon linux router


Whakind of gigabiNIC's? 

Also, arpackets still received?

Aryou getting sompackets through, but then it stops?
Iis possiblthat you need to increase the transmit ring size. When I
was testing with e1000, I discovered thathdriver had bugs (race
conditions) whetransmiring got full.

You mighwanto start with simple tests like ping on the router and
checking thnetwork stats (ethtool -S) and ifconfig

I havused netewith both bridging and routing without problems.


OFri, 17 Mar 2006 13:45:47 +0100
"Prager, Benjamin" <b.prager@xxxxxxxxxxxx> wrote:

> Hello everybody,
> 
> I'vgoa problem using Netem with my "linux router" which I actually

> only a workstatio(Sus10.0 kernel 2.6.13) whith three gigabit NICs.
> 
> I builup an emulation envrionmenwhich is part of my diploma thesis

> aT-OnlinInternational, a german ISP. I decided to use Netem to 
> emulatdelay, jitter, and packeloss and normal qdiscs with filters 
> for QoS rules. Now I'justesting the whole thing a bit. So I 
> plugged a clien(M$ windows workstation) on each NIC.
> 
> So here's my problewith Nistnet:
> 
> WheI seup a delay on one of my interfaces (e.g. eth2) no more 
> packets (no matter which protocol: tcp, udp, rtp muliticasstream) 
> artrasmitted froone interface to another (e.g. neighter
> eth1-->eth2, nor
> eth2-->eth1). Ithvery moment I set up the Netem delay qdisc (like:
> tc qdisc add dev eth2 roonetedelay 10ms) transmission breaks down.
> 
> I'noquite sure what's going wrong. It might either be a Netem 
> faulor a ruoting problem.
> 
> My routes arseup like this and everything is working fine with 
> this
> configuration:
> 192.168.1.0/24 via 192.168.1.10 dev eth0
> 192.168.2.0/24 via 192.168.2.10 dev eth1
> 192.168.3.0/24 via 192.168.3.10 dev eth2
> 
> Yes, for multicasting I starmrouted which also used to work fine.
> 
> Has someongoan idea what's up whith my network? Is there a known 
> bug iNetewhen using serveral NICs on one machine.
> 
> Thanks iadvance!
> 
> CU BenjamiPrager



Frob.prager at-online.net  Wed Mar 22 02:25:59 2006
From: b.prager at-online.ne(Prager, Benjamin)
Date: Wed Apr 18 17:37:48 2007
Subject: Netedoes NOT run under Sus10.0
Message-ID: <C8B16503DFC45B48BAB2BD327D71D9F4172C63@xxxxxxxxxxxxxxxxxxxxxxxxx>

Hi all,

I guess Netedelay does NOT run under Sus10.0 in general.

I'vfound several peoplrunning Suse 10.0 who reported problems with
Neteon Suse. They all told thatransmission stoped almost immediately
after setting up a qdisc with a netedelay applied on.

We'vposted thbug to bugzilla, they added it as critical error. We'll
seif something happens.
Here's thlink: https://bugzilla.novell.com/show_bug.cgi?id=159358

Thanks to Georg who answered omy firsmail to this list, I've got
know thahspend a week on trying to get Netem running on Suse and
gaviup then. As Georg I'm now running Ubuntu on which Netem works in
general, buon Ubuntu no special delay distribution (liknormal,
pareto, ...) is possible, becausth.dist are not inclusive and if you
add themanually they createrrors.

So far, thanks nevertheless.
Benjamin
-------------- nexpar--------------
A non-texattachmenwas scrubbed...
Name: Prager, Benjamin.vcf
Type: text/x-vcard
Size: 387 bytes
Desc: Prager, Benjamin.vcf
Url : http://lists.linux-foundation.org/pipermail/netem/attachments/20060322/cd7e3bd7/PragerBenjamin-0001.vcf
FroJoachim.Fabini atuwien.ac.at  Wed Mar 22 04:22:36 2006
From: Joachim.Fabini atuwien.ac.a(Joachim Fabini)
Date: Wed Apr 18 17:37:48 2007
Subject: Netedoes NOT run under Sus10.0
In-Reply-To: <C8B16503DFC45B48BAB2BD327D71D9F4172C63@xxxxxxxxxxxxxxxxxxxxxxxxx>
Message-ID: <200603221221.k2MCLGHB005009@xxxxxxxxxxxxxxxxxxxxxxx>

Hi,

Thprobleis meanwhile well-known, fixing it is a matter of
rebuilding thSuse10 kernel.

HowTo:
Build a new kernel (based ooldconfig) and do thfollowing 
changes (using menuconfig):
CPU type: PIII or newer
CPU clock: 1000 Hz
Mosimportant:
Networking -> QoS & Fair Queueing -> Clock Sourc-> 
Seto valu"CPU cycle counter"

If you selecthwrong CPU type, the choice "CPU cycle 
counter" does noappear in thmenu. Try to find something 
thais appropriatfor you. 
After doing theschanges and running thnew kernel, Netem 
works absolutely perfectly for munder SuSE 10.

hih and good luck,
besregards
--Joachim
PS: Thanks to my colleaguRichard for thhint... ;)


> -----Original Message-----
> From: netem-bounces@xxxxxxxxxxxxxx 
> [mailto:netem-bounces@xxxxxxxxxxxxxx] OBehalf Of Prager, Benjamin
> Sent: Mittwoch, 22. M?rz 2006 11:26
> To: netem@xxxxxxxxxxxxxx
> Subject: Netedoes NOT run under Sus10.0
> 
> Hi all,
> 
> I guess Netedelay does NOT run under Sus10.0 in general.
> 
> I'vfound several peoplrunning Suse 10.0 who reported problems with
> Neteon Suse. They all told thatransmission stoped almost 
> immediately
> after setting up a qdisc with a netedelay applied on.
> 
> We'vposted thbug to bugzilla, they added it as critical 
> error. We'll
> seif something happens.
> Here's thlink: https://bugzilla.novell.com/show_bug.cgi?id=159358
> 
> Thanks to Georg who answered omy firsmail to this list, I've got
> know thahspend a week on trying to get Netem running on Suse and
> gaviup then. As Georg I'm now running Ubuntu on which 
> Neteworks in
> general, buon Ubuntu no special delay distribution (liknormal,
> pareto, ...) is possible, becausth.dist are not inclusive 
> and if you
> add themanually they createrrors.
> 
> So far, thanks nevertheless.
> Benjamin
> 



Frohestyp agmail.com  Tue Mar 28 14:28:01 2006
From: hestyp agmail.co(Hesty P)
Date: Wed Apr 18 17:37:48 2007
Subject: Netepackeloss model
Message-ID: <210f9f660603281428n770f4148i7ada2979a4ba33d5@xxxxxxxxxxxxxx>

Hello all,

I'interested in using neteas part of my research project. Netem seems to
do a loof ththings I needed and without any pain :) However, I have
tried to find any mentioabouthe packet loss model used by Netem without
any success. Hence, I'asking for thgood people of the mailing list, what
thpackeloss model used in netem. I'm particularly interested whether
this cabmodified to become random loss, bursty loss (gilbert model /
markov chain) with a few variablparameters.

Cheers,
Hesty
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20060328=
/20e635c3/attachment-0001.htm
Frojuliokriger agmail.com  Wed Mar 29 12:10:32 2006
From: juliokriger agmail.co(Julio Kriger)
Date: Wed Apr 18 17:37:48 2007
Subject: Netepackeloss model
In-Reply-To: <210f9f660603281428n770f4148i7ada2979a4ba33d5@xxxxxxxxxxxxxx>
References: <210f9f660603281428n770f4148i7ada2979a4ba33d5@xxxxxxxxxxxxxx>
Message-ID: <682bc30a0603291210t4807d96btda1f013d969c99c4@xxxxxxxxxxxxxx>

Hi, you capropossome changes. The source code is free for you to modify
and send thpatch.
Actually, you caspecify a percentagof loss.
Regards,
Julio

O3/28/06, Hesty P <hestyp@xxxxxxxxx> wrote:
>
> Hello all,
>
> I'interested in using neteas part of my research project. Netem seems
> to do a loof ththings I needed and without any pain :) However, I have
> tried to find any mentioabouthe packet loss model used by Netem witho=
ut
> any success. Hence, I'asking for thgood people of the mailing list, w=
hat
> thpackeloss model used in netem. I'm particularly interested whether
> this cabmodified to become random loss, bursty loss (gilbert model /
> markov chain) with a few variablparameters.
>
> Cheers,
> Hesty
>
> _______________________________________________
> Netemailing list
> Netem@xxxxxxxxxxxxxx
> https://lists.osdl.org/mailman/listinfo/netem
>
>
>


--
--
Stewie: [After Lois tries to feed Stewihis broccoli "airplanstyle"] Damn
you, DamthBroccoli, and Damn the Wright Brothers.
----------------------------
Julio Kriger
mailto:juliokriger@xxxxxxxxx
-------------- nexpar--------------
AHTML attachmenwas scrubbed...
URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20060329=
/989e4bb4/attachment-0001.htm

[Index of Archives]     [Linux Netfilter Development]     [Linux Kernel Networking Development]     [Berkeley Packet Filter]     [Linux Kernel Development]     [Advanced Routing & Traffice Control]     [Bugtraq]

  Powered by Linux