Hi, I wonder: does netework with vlan interfaces? If so I could try setting up onbox plus a switch, giveach pora different vlan and thus have can simulateach poras a modem/gprs/isdn/t1/dsl/cable/... uplink with differensettings. bubeforspending time on developing this, it would breassuring to know if neteand vlan interfaces work together, maybsomeonalready tried that? Regards, Andreas FroCalum.Lind anewport-networks.com Fri Jan 12 06:14:27 2007 From: Calum.Lind anewport-networks.co(Calum Lind) Date: Wed Apr 18 12:51:20 2007 Subject: does netework with vlan? In-Reply-To: <45A76201.3080106@xxxxxxxxxxxxxxx> Message-ID: <004f01c73653$faaf33c0$4301a8c0@xxxxxxxxxxxxxxxxxxx> I aunsuras to what you are trying to achieve with this setup but we use neteboxes sitting between clients and servers as ip forwarders and then seneterules based on ip source and destination addresses. So in your cassimply assign a differenip address (or port) to your different types of connection. I cannosea use for vlans but a little clarification of your setup may provmwrong. Btw this is not to say that vlans won't work with netem, we havsimply never needed iso never tested it. Regards, Calum -----Original Message----- From: netem-bounces@xxxxxxxxxxxxxx [mailto:netem-bounces@xxxxxxxxxxxxxx] On Behalf Of Andreas Jellinghaus Sent: 12 January 2007 10:25 To: netem@xxxxxxxxxxxxxx Subject: does netework with vlan? Hi, I wonder: does netework with vlan interfaces? If so I could try setting up onbox plus a switch, giveach pora different vlan and thus have can simulateach poras a modem/gprs/isdn/t1/dsl/cable/... uplink with differensettings. bubeforspending time on developing this, it would breassuring to know if neteand vlan interfaces work together, maybsomeonalready tried that? Regards, Andreas _______________________________________________ Netemailing list Netem@xxxxxxxxxxxxxx https://lists.osdl.org/mailman/listinfo/netem --------------- This e-mail may contaiconfidential and/or privileged information. If you arnot the intended recipient (or have received this e-mail in error) please notify the sender immediately and delete this e-mail. Any unauthorized copying, disclosure or distribution of the contents in this e-mail is strictly forbidden. --------------- Froshemminger aosdl.org Fri Jan 12 13:30:42 2007 From: shemminger aosdl.org (Stephen Hemminger) Date: Wed Apr 18 12:51:20 2007 Subject: does netework with vlan? In-Reply-To: <45A76201.3080106@xxxxxxxxxxxxxxx> References: <45A76201.3080106@xxxxxxxxxxxxxxx> Message-ID: <20070112133042.450972bd@localhost> OFri, 12 Jan 2007 11:25:05 +0100 Andreas Jellinghaus <aj@xxxxxxxxxxxxxxx> wrote: > Hi, > > I wonder: does netework with vlan interfaces? > > If so I could try setting up onbox plus a switch, > giveach pora different vlan and thus have can > simulateach poras a modem/gprs/isdn/t1/dsl/cable/... > uplink with differensettings. > > bubeforspending time on developing this, it would > breassuring to know if neteand vlan interfaces > work together, maybsomeonalready tried that? > > Regards, Andreas > _______________________________________________ Each vladoes nohave it's own transmit queue so netewon'work as expected. You might make it work with somhacking. Frobrown acis.udel.edu Fri Jan 12 11:04:40 2007 From: browacis.udel.edu (Aaron Brown) Date: Wed Apr 18 12:51:20 2007 Subject: gigabispeeds Message-ID: <45A7DBC8.5000708@xxxxxxxxxxxx> I'vsetup a network with a topology of a-b-c-d. Thmachines have 1 Pentiu4 2.8GHz, 4 Gigs of RAM, e1000 NICs and arrunning Fedora Core 4(2.6.17-1.2142_FC4smp). Withousetting up delays, I gespeeds of ~945Mbit/s using iperf iUDP mode. However, oncI add delays in, the speed starts dropping: 15ms delay ieach direction between b-c: 775 Mbit/s 25ms delay ieach direction between b-c: 462 Mbit/s 50ms delay ieach direction between b-c: 226 Mbit/s Army machines nopowerful enough to use netem at GigE speeds or have I simply misconfigured something? Thanks. Cheers, Aaron Frocalum.lind anewport-networks.com Fri Jan 12 13:53:25 2007 From: calum.lind anewport-networks.co(Calum Lind) Date: Wed Apr 18 12:51:20 2007 Subject: gigabispeeds In-Reply-To: <45A7DBC8.5000708@xxxxxxxxxxxx> Message-ID: <000001c73694$187acff0$5601a8c0@xxxxxxxxxxxxxxxxxxx> I think Stephewould bmore qualified to answer this question but my educated guess would bthait is to do with the buffer size, which needs increasing. Basically thhigher thdelay the greater the number packets your machinhas to storbefore forwarding on. Btw simply monitoring cpu usagwill leyou know if it is actually the machinthais the bottleneck. -----Original Message----- From: netem-bounces@xxxxxxxxxxxxxx [mailto:netem-bounces@xxxxxxxxxxxxxx] On Behalf Of AaroBrown Sent: 12 January 2007 19:05 To: netem@xxxxxxxxxxxxxx Subject: gigabispeeds I'vsetup a network with a topology of a-b-c-d. Thmachines have 1 Pentiu4 2.8GHz, 4 Gigs of RAM, e1000 NICs and arrunning Fedora Core 4(2.6.17-1.2142_FC4smp). Withousetting up delays, I gespeeds of ~945Mbit/s using iperf iUDP mode. However, oncI add delays in, the speed starts dropping: 15ms delay ieach direction between b-c: 775 Mbit/s 25ms delay ieach direction between b-c: 462 Mbit/s 50ms delay ieach direction between b-c: 226 Mbit/s Army machines nopowerful enough to use netem at GigE speeds or have I simply misconfigured something? Thanks. Cheers, Aaron _______________________________________________ Netemailing list Netem@xxxxxxxxxxxxxx https://lists.osdl.org/mailman/listinfo/netem --------------- This e-mail may contaiconfidential and/or privileged information. If you arnot the intended recipient (or have received this e-mail in error) please notify the sender immediately and delete this e-mail. Any unauthorized copying, disclosure or distribution of the contents in this e-mail is strictly forbidden. --------------- Froshemminger aosdl.org Fri Jan 12 14:20:25 2007 From: shemminger aosdl.org (Stephen Hemminger) Date: Wed Apr 18 12:51:21 2007 Subject: gigabispeeds In-Reply-To: <000001c73694$187acff0$5601a8c0@xxxxxxxxxxxxxxxxxxx> References: <45A7DBC8.5000708@xxxxxxxxxxxx> <000001c73694$187acff0$5601a8c0@xxxxxxxxxxxxxxxxxxx> Message-ID: <20070112142025.7baaf14b@localhost> OFri, 12 Jan 2007 21:53:25 -0000 "CaluLind" <calum.lind@xxxxxxxxxxxxxxxxxxxx> wrote: > I think Stephewould bmore qualified to answer this question but my > educated guess would bthait is to do with the buffer size, which needs > increasing. Basically thhigher thdelay the greater the number packets > your machinhas to storbefore forwarding on. > > Btw simply monitoring cpu usagwill leyou know if it is actually the > machinthais the bottleneck. > > -----Original Message----- > From: netem-bounces@xxxxxxxxxxxxxx [mailto:netem-bounces@xxxxxxxxxxxxxx] On > Behalf Of AaroBrown > Sent: 12 January 2007 19:05 > To: netem@xxxxxxxxxxxxxx > Subject: gigabispeeds > > I'vsetup a network with a topology of a-b-c-d. Thmachines have 1 > Pentiu4 2.8GHz, 4 Gigs of RAM, e1000 NICs and arrunning Fedora Core > 4(2.6.17-1.2142_FC4smp). Withousetting up delays, I gespeeds of > ~945Mbit/s using iperf iUDP mode. However, oncI add delays in, the > speed starts dropping: > > 15ms delay ieach direction between b-c: 775 Mbit/s > 25ms delay ieach direction between b-c: 462 Mbit/s > 50ms delay ieach direction between b-c: 226 Mbit/s > > Army machines nopowerful enough to use netem at GigE speeds or have > I simply misconfigured something? Thanks. > > Cheers, > Aaron iperf UDP tesuses ratrestriction (in application). What do you see with TCP. Also does iperf reporany packeloss when running over UDP? Another factor is thaUDP sends lots of small packets, vs TCP which sends thlargespossible. Frobrown acis.udel.edu Sat Jan 13 08:30:45 2007 From: browacis.udel.edu (Aaron Brown) Date: Wed Apr 18 12:51:21 2007 Subject: gigabispeeds In-Reply-To: <20070112142025.7baaf14b@localhost> References: <45A7DBC8.5000708@xxxxxxxxxxxx> <000001c73694$187acff0$5601a8c0@xxxxxxxxxxxxxxxxxxx> <20070112142025.7baaf14b@localhost> Message-ID: <1F2F06C7-5DF7-4F95-B402-8C69FF04607B@xxxxxxxxxxxx> OJan 12, 2007, a4:53 PM, Calum Lind wrote: > I think Stephewould bmore qualified to answer this question but my > educated guess would bthait is to do with the buffer size, > which needs > increasing. Basically thhigher thdelay the greater the number > packets > your machinhas to storbefore forwarding on. > > Btw simply monitoring cpu usagwill leyou know if it is actually > the > machinthais the bottleneck. Thmachindoesn't appear to be the bottleneck, it's running at 25% cpu utilization. Which buffers need to bincreased? OJan 12, 2007, a5:20 PM, Stephen Hemminger wrote: > iperf UDP tesuses ratrestriction (in application). What do you > sewith > TCP. Also does iperf reporany packeloss when running over UDP? UDP's outgoing ratwas seto 1Gb/s and it was able to send ~970Mb/ s, but, a loof packets werlost. According to the receiving side, abou3/4 of thpackets are dropped with 100ms delay. TCP: 15 ms delay: 405 Mb/s 25 ms delay: 218 Mb/s 50 ms delay: 110 Mb/s Thanks Cheers, Aaron Frobaumann atik.ee.ethz.ch Sun Jan 14 13:40:18 2007 From: baumanatik.ee.ethz.ch (Rainer Baumann) Date: Wed Apr 18 12:51:21 2007 Subject: [Fwd: [PATCH 2.6.18 0/2] LARTC: traccontrol for netem] In-Reply-To: <45827B5C.3090402@xxxxxxxxxxxxxx> References: <45827B5C.3090402@xxxxxxxxxxxxxx> Message-ID: <45AAA342.4070208@xxxxxxxxxxxxxx> Hi Stephen I juswanted to ask you, if you already had timto test our trace extensiofor neteas discussed on the 13th of December. Cheers Rainer Rainer Baumanwrote: > Hi Stephen > > As discussed yesterday, herour patches to integrattrace control into netem > > > > TracControl for Netem: Emulatnetwork properties such as long range dependency and self-similarity of cross-traffic. > > A new optio(trace) has been added to thnetem command. If the trace option is used, the values for packet delay etc. are read from a pregenerated trace file, afterwards the packets are processed by the normal netem functions. The packet action values are readout from the trace file in user space and sent to kernel space via configfs. > > > > > > > > _______________________________________________ > Netemailing list > Netem@xxxxxxxxxxxxxx > https://lists.osdl.org/mailman/listinfo/netem > Frolaurent.mariat thomson.net Mon Jan 15 02:48:56 2007 From: laurent.mariathomson.net (Marie Laurent) Date: Wed Apr 18 12:51:21 2007 Subject: gigabispeeds Message-ID: <E55CA8ECFBA85E4CB8E8173566B9EFEB35CBF4@xxxxxxxxxxxxxxxxxxxxxxxxxx> Hello, Well, Whavbeen using netem to add fixed and random delay on high speed strems oFC4 linux PCs. Typically wforward, 24 hours a day, 400Mb/s to 500Mb/s froone Etherneporto the other. Thbasic limitation is thPCI bus of your PC, not the CPU. Our PCI bus only suppor1.2 to 1.4Gb/s network traffic i.e. 600 to 700Mb/s in + out. Whad to usthe "limit" parameter of the netem traffic controller to increasthnumber of buffers otherwise packets were dropped. The limit parameter is nomentioned on thnetem web site but in tc documentation. It's quitan importanparameter and it would be worth detailed othnetem page (http://linux-net.osdl.org/index.php/Netem). Thlimiwould be greater than the average delay multiplied by the packerate. Typically, wadd 100ms fixed delay +/- 30ms jitter: tc qdisc add dev eth1 roonetelimit 16000 delay 100ms 30ms Yours sincerely, LaurenMARIE -----Original Message----- From: netem-bounces@xxxxxxxxxxxxxx [mailto:netem-bounces@xxxxxxxxxxxxxx] OBehalf Of Aaron Brown Sent: samedi 13 janvier 2007 17:31 To: StepheHemminger Cc: netem@xxxxxxxxxxxxxx Subject: Re: gigabispeeds OJan 12, 2007, a4:53 PM, Calum Lind wrote: > I think Stephewould bmore qualified to answer this question but my > educated guess would bthait is to do with the buffer size, > which needs > increasing. Basically thhigher thdelay the greater the number > packets > your machinhas to storbefore forwarding on. > > Btw simply monitoring cpu usagwill leyou know if it is actually > the > machinthais the bottleneck. Thmachindoesn't appear to be the bottleneck, it's running at 25% cpu utilization. Which buffers need to bincreased? OJan 12, 2007, a5:20 PM, Stephen Hemminger wrote: > iperf UDP tesuses ratrestriction (in application). What do you > sewith > TCP. Also does iperf reporany packeloss when running over UDP? UDP's outgoing ratwas seto 1Gb/s and it was able to send ~970Mb/ s, but, a loof packets werlost. According to the receiving side, abou3/4 of thpackets are dropped with 100ms delay. TCP: 15 ms delay: 405 Mb/s 25 ms delay: 218 Mb/s 50 ms delay: 110 Mb/s Thanks Cheers, Aaron _______________________________________________ Netemailing list Netem@xxxxxxxxxxxxxx https://lists.osdl.org/mailman/listinfo/netem Froaj aciphirelabs.com Mon Jan 15 03:55:39 2007 From: aj aciphirelabs.co(Andreas Jellinghaus) Date: Wed Apr 18 12:51:21 2007 Subject: does netework with vlan? In-Reply-To: <20070112133042.450972bd@localhost> References: <45A76201.3080106@xxxxxxxxxxxxxxx> <20070112133042.450972bd@localhost> Message-ID: <45AB6BBB.9080504@xxxxxxxxxxxxxxx> StepheHemminger wrote: > Each vladoes nohave it's own transmit queue so > netewon'work as expected. You might make it work with > somhacking. ah, I feared something likthat. ok, thanks a lot. :( any idea whawould bnecessary to hack this to get it working? buI don'know kernel programming well, so I will not likely bablto do that. instead I can try putting lots of 4port network cards into onmachine. Regards, Andreas Frobrown acis.udel.edu Mon Jan 15 14:31:31 2007 From: browacis.udel.edu (Aaron Brown) Date: Wed Apr 18 12:51:21 2007 Subject: gigabispeeds In-Reply-To: <E55CA8ECFBA85E4CB8E8173566B9EFEB35CBF4@xxxxxxxxxxxxxxxxxxxxxxxxxx> References: <E55CA8ECFBA85E4CB8E8173566B9EFEB35CBF4@xxxxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <BC21BE85-1D86-4B90-BBA1-CAA03D513E14@xxxxxxxxxxxx> OJan 15, 2007, a5:48 AM, Marie Laurent wrote: > > Whad to usthe "limit" parameter of the netem traffic controller to > increasthnumber of buffers otherwise packets were dropped. Thasolved thproblem: UDP now runs at ~840 Mbit/s with no packet loss. Thanks to everyonwho helped :-) Cheers, Aaron Frobaumann atik.ee.ethz.ch Sat Jan 20 01:05:02 2007 From: baumanatik.ee.ethz.ch (Rainer Baumann) Date: Wed Apr 18 12:51:21 2007 Subject: TCiKern In-Reply-To: <20061213101656.6068c793@xxxxxxxxxxxxxxxxx> References: <4514DC9A.2000505@xxxxxxxxxxxxxx> <20060925132800.09856e10@xxxxxxxxxxxxxxxxx> <45198AF5.9090909@xxxxxxxxxxxxxx> <20060926134531.3ec4991a@freekitty> <457A7F43.3020004@xxxxxxxxxxxxxx> <20061213101656.6068c793@xxxxxxxxxxxxxxxxx> Message-ID: <45B1DB3E.2000109@xxxxxxxxxxxxxx> Hi Stephen Ithlast two weeks i've got 13 request, in which release the trace extensioTCfor netem will be included. Did you already found time to integratit? Cheers Rainer Froshemminger aosdl.org Sat Jan 20 14:18:00 2007 From: shemminger aosdl.org (Stephen Hemminger) Date: Wed Apr 18 12:51:21 2007 Subject: TCiKern In-Reply-To: <45B1DB3E.2000109@xxxxxxxxxxxxxx> References: <4514DC9A.2000505@xxxxxxxxxxxxxx> <20060925132800.09856e10@xxxxxxxxxxxxxxxxx> <45198AF5.9090909@xxxxxxxxxxxxxx> <20060926134531.3ec4991a@freekitty> <457A7F43.3020004@xxxxxxxxxxxxxx> <20061213101656.6068c793@xxxxxxxxxxxxxxxxx> <45B1DB3E.2000109@xxxxxxxxxxxxxx> Message-ID: <20070120221800.7c94d17a@oldman> OSat, 20 Jan 2007 10:05:02 +0100 Rainer Bauman<baumann@xxxxxxxxxxxxxx> wrote: > Hi Stephen > > Ithlast two weeks i've got 13 request, in which release the trace > extensioTCfor netem will be included. Did you already found time to > integratit? > > Cheers > Rainer WheI geback from London, early next week Frotaankr aaston.ac.uk Tue Jan 23 11:21:02 2007 From: taankr aaston.ac.uk (Ritesh Taank) Date: Wed Apr 18 12:51:21 2007 Subject: netepackeloss issue Message-ID: <1169580062.11001.6.camel@localhost.localdomain> Has anybody beeusing th'loss' feature of netem extensively and discovered unusual behaviour? I hava bridgconfigured with two interfaces, where i am able to apply neterules to each interfacto control passing through traffic. Wheusing thcommand: tc qdisc add dev eth0 rooneteloss 0% I agetting unexpected delays in my TCP experiments? A 100MB transfer is taking well over 60 seconds, and whei execut(i.e. taking away the filter completely): tc qdisc del dev eth0 rooneteloss 0% th100MB transfer takes jus8 seconds, which is obviously normal. So whaexactly is going on with thtc filters? Surely setting the loss parameter to 0% should noaffecmy results? Any lighon this would help mwith future experiments. Regards, Ritesh --- AdaptivCommunications Networks Research Group Electronic Engineering Dept. AstoUniversity Birmingham B7 4ET t: +44 (0)7732 069 667 e: taankr@xxxxxxxxxxx -------------- nexpar-------------- AHTML attachmenwas scrubbed... URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20070123/410816cc/attachment.htm Froaj aciphirelabs.com Fri Jan 12 02:25:05 2007 From: aj aciphirelabs.co(Andreas Jellinghaus) Date: Wed Apr 18 17:37:50 2007 Subject: does netework with vlan? Message-ID: <45A76201.3080106@xxxxxxxxxxxxxxx> Hi, I wonder: does netework with vlan interfaces? If so I could try setting up onbox plus a switch, giveach pora different vlan and thus have can simulateach poras a modem/gprs/isdn/t1/dsl/cable/... uplink with differensettings. bubeforspending time on developing this, it would breassuring to know if neteand vlan interfaces work together, maybsomeonalready tried that? Regards, Andreas FroCalum.Lind anewport-networks.com Fri Jan 12 06:14:27 2007 From: Calum.Lind anewport-networks.co(Calum Lind) Date: Wed Apr 18 17:37:50 2007 Subject: does netework with vlan? In-Reply-To: <45A76201.3080106@xxxxxxxxxxxxxxx> Message-ID: <004f01c73653$faaf33c0$4301a8c0@xxxxxxxxxxxxxxxxxxx> I aunsuras to what you are trying to achieve with this setup but we use neteboxes sitting between clients and servers as ip forwarders and then seneterules based on ip source and destination addresses. So in your cassimply assign a differenip address (or port) to your different types of connection. I cannosea use for vlans but a little clarification of your setup may provmwrong. Btw this is not to say that vlans won't work with netem, we havsimply never needed iso never tested it. Regards, Calum -----Original Message----- From: netem-bounces@xxxxxxxxxxxxxx [mailto:netem-bounces@xxxxxxxxxxxxxx] On Behalf Of Andreas Jellinghaus Sent: 12 January 2007 10:25 To: netem@xxxxxxxxxxxxxx Subject: does netework with vlan? Hi, I wonder: does netework with vlan interfaces? If so I could try setting up onbox plus a switch, giveach pora different vlan and thus have can simulateach poras a modem/gprs/isdn/t1/dsl/cable/... uplink with differensettings. bubeforspending time on developing this, it would breassuring to know if neteand vlan interfaces work together, maybsomeonalready tried that? Regards, Andreas _______________________________________________ Netemailing list Netem@xxxxxxxxxxxxxx https://lists.osdl.org/mailman/listinfo/netem --------------- This e-mail may contaiconfidential and/or privileged information. If you arnot the intended recipient (or have received this e-mail in error) please notify the sender immediately and delete this e-mail. Any unauthorized copying, disclosure or distribution of the contents in this e-mail is strictly forbidden. --------------- Froshemminger aosdl.org Fri Jan 12 13:30:42 2007 From: shemminger aosdl.org (Stephen Hemminger) Date: Wed Apr 18 17:37:50 2007 Subject: does netework with vlan? In-Reply-To: <45A76201.3080106@xxxxxxxxxxxxxxx> References: <45A76201.3080106@xxxxxxxxxxxxxxx> Message-ID: <20070112133042.450972bd@localhost> OFri, 12 Jan 2007 11:25:05 +0100 Andreas Jellinghaus <aj@xxxxxxxxxxxxxxx> wrote: > Hi, > > I wonder: does netework with vlan interfaces? > > If so I could try setting up onbox plus a switch, > giveach pora different vlan and thus have can > simulateach poras a modem/gprs/isdn/t1/dsl/cable/... > uplink with differensettings. > > bubeforspending time on developing this, it would > breassuring to know if neteand vlan interfaces > work together, maybsomeonalready tried that? > > Regards, Andreas > _______________________________________________ Each vladoes nohave it's own transmit queue so netewon'work as expected. You might make it work with somhacking. Frobrown acis.udel.edu Fri Jan 12 11:04:40 2007 From: browacis.udel.edu (Aaron Brown) Date: Wed Apr 18 17:37:50 2007 Subject: gigabispeeds Message-ID: <45A7DBC8.5000708@xxxxxxxxxxxx> I'vsetup a network with a topology of a-b-c-d. Thmachines have 1 Pentiu4 2.8GHz, 4 Gigs of RAM, e1000 NICs and arrunning Fedora Core 4(2.6.17-1.2142_FC4smp). Withousetting up delays, I gespeeds of ~945Mbit/s using iperf iUDP mode. However, oncI add delays in, the speed starts dropping: 15ms delay ieach direction between b-c: 775 Mbit/s 25ms delay ieach direction between b-c: 462 Mbit/s 50ms delay ieach direction between b-c: 226 Mbit/s Army machines nopowerful enough to use netem at GigE speeds or have I simply misconfigured something? Thanks. Cheers, Aaron Frocalum.lind anewport-networks.com Fri Jan 12 13:53:25 2007 From: calum.lind anewport-networks.co(Calum Lind) Date: Wed Apr 18 17:37:50 2007 Subject: gigabispeeds In-Reply-To: <45A7DBC8.5000708@xxxxxxxxxxxx> Message-ID: <000001c73694$187acff0$5601a8c0@xxxxxxxxxxxxxxxxxxx> I think Stephewould bmore qualified to answer this question but my educated guess would bthait is to do with the buffer size, which needs increasing. Basically thhigher thdelay the greater the number packets your machinhas to storbefore forwarding on. Btw simply monitoring cpu usagwill leyou know if it is actually the machinthais the bottleneck. -----Original Message----- From: netem-bounces@xxxxxxxxxxxxxx [mailto:netem-bounces@xxxxxxxxxxxxxx] On Behalf Of AaroBrown Sent: 12 January 2007 19:05 To: netem@xxxxxxxxxxxxxx Subject: gigabispeeds I'vsetup a network with a topology of a-b-c-d. Thmachines have 1 Pentiu4 2.8GHz, 4 Gigs of RAM, e1000 NICs and arrunning Fedora Core 4(2.6.17-1.2142_FC4smp). Withousetting up delays, I gespeeds of ~945Mbit/s using iperf iUDP mode. However, oncI add delays in, the speed starts dropping: 15ms delay ieach direction between b-c: 775 Mbit/s 25ms delay ieach direction between b-c: 462 Mbit/s 50ms delay ieach direction between b-c: 226 Mbit/s Army machines nopowerful enough to use netem at GigE speeds or have I simply misconfigured something? Thanks. Cheers, Aaron _______________________________________________ Netemailing list Netem@xxxxxxxxxxxxxx https://lists.osdl.org/mailman/listinfo/netem --------------- This e-mail may contaiconfidential and/or privileged information. If you arnot the intended recipient (or have received this e-mail in error) please notify the sender immediately and delete this e-mail. Any unauthorized copying, disclosure or distribution of the contents in this e-mail is strictly forbidden. --------------- Froshemminger aosdl.org Fri Jan 12 14:20:25 2007 From: shemminger aosdl.org (Stephen Hemminger) Date: Wed Apr 18 17:37:50 2007 Subject: gigabispeeds In-Reply-To: <000001c73694$187acff0$5601a8c0@xxxxxxxxxxxxxxxxxxx> References: <45A7DBC8.5000708@xxxxxxxxxxxx> <000001c73694$187acff0$5601a8c0@xxxxxxxxxxxxxxxxxxx> Message-ID: <20070112142025.7baaf14b@localhost> OFri, 12 Jan 2007 21:53:25 -0000 "CaluLind" <calum.lind@xxxxxxxxxxxxxxxxxxxx> wrote: > I think Stephewould bmore qualified to answer this question but my > educated guess would bthait is to do with the buffer size, which needs > increasing. Basically thhigher thdelay the greater the number packets > your machinhas to storbefore forwarding on. > > Btw simply monitoring cpu usagwill leyou know if it is actually the > machinthais the bottleneck. > > -----Original Message----- > From: netem-bounces@xxxxxxxxxxxxxx [mailto:netem-bounces@xxxxxxxxxxxxxx] On > Behalf Of AaroBrown > Sent: 12 January 2007 19:05 > To: netem@xxxxxxxxxxxxxx > Subject: gigabispeeds > > I'vsetup a network with a topology of a-b-c-d. Thmachines have 1 > Pentiu4 2.8GHz, 4 Gigs of RAM, e1000 NICs and arrunning Fedora Core > 4(2.6.17-1.2142_FC4smp). Withousetting up delays, I gespeeds of > ~945Mbit/s using iperf iUDP mode. However, oncI add delays in, the > speed starts dropping: > > 15ms delay ieach direction between b-c: 775 Mbit/s > 25ms delay ieach direction between b-c: 462 Mbit/s > 50ms delay ieach direction between b-c: 226 Mbit/s > > Army machines nopowerful enough to use netem at GigE speeds or have > I simply misconfigured something? Thanks. > > Cheers, > Aaron iperf UDP tesuses ratrestriction (in application). What do you see with TCP. Also does iperf reporany packeloss when running over UDP? Another factor is thaUDP sends lots of small packets, vs TCP which sends thlargespossible. Frobrown acis.udel.edu Sat Jan 13 08:30:45 2007 From: browacis.udel.edu (Aaron Brown) Date: Wed Apr 18 17:37:50 2007 Subject: gigabispeeds In-Reply-To: <20070112142025.7baaf14b@localhost> References: <45A7DBC8.5000708@xxxxxxxxxxxx> <000001c73694$187acff0$5601a8c0@xxxxxxxxxxxxxxxxxxx> <20070112142025.7baaf14b@localhost> Message-ID: <1F2F06C7-5DF7-4F95-B402-8C69FF04607B@xxxxxxxxxxxx> OJan 12, 2007, a4:53 PM, Calum Lind wrote: > I think Stephewould bmore qualified to answer this question but my > educated guess would bthait is to do with the buffer size, > which needs > increasing. Basically thhigher thdelay the greater the number > packets > your machinhas to storbefore forwarding on. > > Btw simply monitoring cpu usagwill leyou know if it is actually > the > machinthais the bottleneck. Thmachindoesn't appear to be the bottleneck, it's running at 25% cpu utilization. Which buffers need to bincreased? OJan 12, 2007, a5:20 PM, Stephen Hemminger wrote: > iperf UDP tesuses ratrestriction (in application). What do you > sewith > TCP. Also does iperf reporany packeloss when running over UDP? UDP's outgoing ratwas seto 1Gb/s and it was able to send ~970Mb/ s, but, a loof packets werlost. According to the receiving side, abou3/4 of thpackets are dropped with 100ms delay. TCP: 15 ms delay: 405 Mb/s 25 ms delay: 218 Mb/s 50 ms delay: 110 Mb/s Thanks Cheers, Aaron Frobaumann atik.ee.ethz.ch Sun Jan 14 13:40:18 2007 From: baumanatik.ee.ethz.ch (Rainer Baumann) Date: Wed Apr 18 17:37:50 2007 Subject: [Fwd: [PATCH 2.6.18 0/2] LARTC: traccontrol for netem] In-Reply-To: <45827B5C.3090402@xxxxxxxxxxxxxx> References: <45827B5C.3090402@xxxxxxxxxxxxxx> Message-ID: <45AAA342.4070208@xxxxxxxxxxxxxx> Hi Stephen I juswanted to ask you, if you already had timto test our trace extensiofor neteas discussed on the 13th of December. Cheers Rainer Rainer Baumanwrote: > Hi Stephen > > As discussed yesterday, herour patches to integrattrace control into netem > > > > TracControl for Netem: Emulatnetwork properties such as long range dependency and self-similarity of cross-traffic. > > A new optio(trace) has been added to thnetem command. If the trace option is used, the values for packet delay etc. are read from a pregenerated trace file, afterwards the packets are processed by the normal netem functions. The packet action values are readout from the trace file in user space and sent to kernel space via configfs. > > > > > > > > _______________________________________________ > Netemailing list > Netem@xxxxxxxxxxxxxx > https://lists.osdl.org/mailman/listinfo/netem > Frolaurent.mariat thomson.net Mon Jan 15 02:48:56 2007 From: laurent.mariathomson.net (Marie Laurent) Date: Wed Apr 18 17:37:50 2007 Subject: gigabispeeds Message-ID: <E55CA8ECFBA85E4CB8E8173566B9EFEB35CBF4@xxxxxxxxxxxxxxxxxxxxxxxxxx> Hello, Well, Whavbeen using netem to add fixed and random delay on high speed strems oFC4 linux PCs. Typically wforward, 24 hours a day, 400Mb/s to 500Mb/s froone Etherneporto the other. Thbasic limitation is thPCI bus of your PC, not the CPU. Our PCI bus only suppor1.2 to 1.4Gb/s network traffic i.e. 600 to 700Mb/s in + out. Whad to usthe "limit" parameter of the netem traffic controller to increasthnumber of buffers otherwise packets were dropped. The limit parameter is nomentioned on thnetem web site but in tc documentation. It's quitan importanparameter and it would be worth detailed othnetem page (http://linux-net.osdl.org/index.php/Netem). Thlimiwould be greater than the average delay multiplied by the packerate. Typically, wadd 100ms fixed delay +/- 30ms jitter: tc qdisc add dev eth1 roonetelimit 16000 delay 100ms 30ms Yours sincerely, LaurenMARIE -----Original Message----- From: netem-bounces@xxxxxxxxxxxxxx [mailto:netem-bounces@xxxxxxxxxxxxxx] OBehalf Of Aaron Brown Sent: samedi 13 janvier 2007 17:31 To: StepheHemminger Cc: netem@xxxxxxxxxxxxxx Subject: Re: gigabispeeds OJan 12, 2007, a4:53 PM, Calum Lind wrote: > I think Stephewould bmore qualified to answer this question but my > educated guess would bthait is to do with the buffer size, > which needs > increasing. Basically thhigher thdelay the greater the number > packets > your machinhas to storbefore forwarding on. > > Btw simply monitoring cpu usagwill leyou know if it is actually > the > machinthais the bottleneck. Thmachindoesn't appear to be the bottleneck, it's running at 25% cpu utilization. Which buffers need to bincreased? OJan 12, 2007, a5:20 PM, Stephen Hemminger wrote: > iperf UDP tesuses ratrestriction (in application). What do you > sewith > TCP. Also does iperf reporany packeloss when running over UDP? UDP's outgoing ratwas seto 1Gb/s and it was able to send ~970Mb/ s, but, a loof packets werlost. According to the receiving side, abou3/4 of thpackets are dropped with 100ms delay. TCP: 15 ms delay: 405 Mb/s 25 ms delay: 218 Mb/s 50 ms delay: 110 Mb/s Thanks Cheers, Aaron _______________________________________________ Netemailing list Netem@xxxxxxxxxxxxxx https://lists.osdl.org/mailman/listinfo/netem Froaj aciphirelabs.com Mon Jan 15 03:55:39 2007 From: aj aciphirelabs.co(Andreas Jellinghaus) Date: Wed Apr 18 17:37:50 2007 Subject: does netework with vlan? In-Reply-To: <20070112133042.450972bd@localhost> References: <45A76201.3080106@xxxxxxxxxxxxxxx> <20070112133042.450972bd@localhost> Message-ID: <45AB6BBB.9080504@xxxxxxxxxxxxxxx> StepheHemminger wrote: > Each vladoes nohave it's own transmit queue so > netewon'work as expected. You might make it work with > somhacking. ah, I feared something likthat. ok, thanks a lot. :( any idea whawould bnecessary to hack this to get it working? buI don'know kernel programming well, so I will not likely bablto do that. instead I can try putting lots of 4port network cards into onmachine. Regards, Andreas Frobrown acis.udel.edu Mon Jan 15 14:31:31 2007 From: browacis.udel.edu (Aaron Brown) Date: Wed Apr 18 17:37:50 2007 Subject: gigabispeeds In-Reply-To: <E55CA8ECFBA85E4CB8E8173566B9EFEB35CBF4@xxxxxxxxxxxxxxxxxxxxxxxxxx> References: <E55CA8ECFBA85E4CB8E8173566B9EFEB35CBF4@xxxxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <BC21BE85-1D86-4B90-BBA1-CAA03D513E14@xxxxxxxxxxxx> OJan 15, 2007, a5:48 AM, Marie Laurent wrote: > > Whad to usthe "limit" parameter of the netem traffic controller to > increasthnumber of buffers otherwise packets were dropped. Thasolved thproblem: UDP now runs at ~840 Mbit/s with no packet loss. Thanks to everyonwho helped :-) Cheers, Aaron Frobaumann atik.ee.ethz.ch Sat Jan 20 01:05:02 2007 From: baumanatik.ee.ethz.ch (Rainer Baumann) Date: Wed Apr 18 17:37:50 2007 Subject: TCiKern In-Reply-To: <20061213101656.6068c793@xxxxxxxxxxxxxxxxx> References: <4514DC9A.2000505@xxxxxxxxxxxxxx> <20060925132800.09856e10@xxxxxxxxxxxxxxxxx> <45198AF5.9090909@xxxxxxxxxxxxxx> <20060926134531.3ec4991a@freekitty> <457A7F43.3020004@xxxxxxxxxxxxxx> <20061213101656.6068c793@xxxxxxxxxxxxxxxxx> Message-ID: <45B1DB3E.2000109@xxxxxxxxxxxxxx> Hi Stephen Ithlast two weeks i've got 13 request, in which release the trace extensioTCfor netem will be included. Did you already found time to integratit? Cheers Rainer Froshemminger aosdl.org Sat Jan 20 14:18:00 2007 From: shemminger aosdl.org (Stephen Hemminger) Date: Wed Apr 18 17:37:50 2007 Subject: TCiKern In-Reply-To: <45B1DB3E.2000109@xxxxxxxxxxxxxx> References: <4514DC9A.2000505@xxxxxxxxxxxxxx> <20060925132800.09856e10@xxxxxxxxxxxxxxxxx> <45198AF5.9090909@xxxxxxxxxxxxxx> <20060926134531.3ec4991a@freekitty> <457A7F43.3020004@xxxxxxxxxxxxxx> <20061213101656.6068c793@xxxxxxxxxxxxxxxxx> <45B1DB3E.2000109@xxxxxxxxxxxxxx> Message-ID: <20070120221800.7c94d17a@oldman> OSat, 20 Jan 2007 10:05:02 +0100 Rainer Bauman<baumann@xxxxxxxxxxxxxx> wrote: > Hi Stephen > > Ithlast two weeks i've got 13 request, in which release the trace > extensioTCfor netem will be included. Did you already found time to > integratit? > > Cheers > Rainer WheI geback from London, early next week Frotaankr aaston.ac.uk Tue Jan 23 11:21:02 2007 From: taankr aaston.ac.uk (Ritesh Taank) Date: Wed Apr 18 17:37:50 2007 Subject: netepackeloss issue Message-ID: <1169580062.11001.6.camel@localhost.localdomain> Has anybody beeusing th'loss' feature of netem extensively and discovered unusual behaviour? I hava bridgconfigured with two interfaces, where i am able to apply neterules to each interfacto control passing through traffic. Wheusing thcommand: tc qdisc add dev eth0 rooneteloss 0% I agetting unexpected delays in my TCP experiments? A 100MB transfer is taking well over 60 seconds, and whei execut(i.e. taking away the filter completely): tc qdisc del dev eth0 rooneteloss 0% th100MB transfer takes jus8 seconds, which is obviously normal. So whaexactly is going on with thtc filters? Surely setting the loss parameter to 0% should noaffecmy results? Any lighon this would help mwith future experiments. Regards, Ritesh --- AdaptivCommunications Networks Research Group Electronic Engineering Dept. AstoUniversity Birmingham B7 4ET t: +44 (0)7732 069 667 e: taankr@xxxxxxxxxxx -------------- nexpar-------------- AHTML attachmenwas scrubbed... URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20070123/410816cc/attachment-0001.htm
