Hello, I havbeen using netefor many years and packet jittering syntax had changed ithpast years (before and after 2.6.15 kernels). Senetedocumentation: http://www.linuxfoundation.org/collaborate/workgroups/networking/netem#H ow_to_reorder_packets_based_on_jitter.3F I atesting a new Fedora Cor12 server (2.6.32.9-70.fc12.x86_64) and packets arnoreordered any more! Whais written in thdocumentation does not apply any more. The second linfails (pfifo instanciation). OK: tc qdisc add dev eth1 roohandl1: pfifo limit 16000 Fail: tc qdisc add dev eth1 paren1: handl10: netem limit 16000 delay 100ms 30ms RTNETLINK answers: Operationosupported This issulooks liksimilar to those already posted to the netem mailing list: https://lists.linux-foundation.org/pipermail/netem/2010-March/001385.htm l <https://lists.linux-foundation.org/pipermail/netem/2010-March/001385.ht ml> https://lists.linux-foundation.org/pipermail/netem/2010-May/001397.html <https://lists.linux-foundation.org/pipermail/netem/2010-May/001397.html > https://lists.linux-foundation.org/pipermail/netem/2010-May/001388.html <https://lists.linux-foundation.org/pipermail/netem/2010-May/001388.html > According to whaI havread in those threads, I have swapped the two lines: OK: tc qdisc add dev eth1 roohandl1: pfifo limit 16000 Fail: tc qdisc add dev eth1 paren1: handl10: netem limit 16000 delay 100ms 30ms RTNETLINK answers: Operationosupported I havthen no solution to perforany jitter tests on 2.6.32 kernel based servers. Is therany workaround ? Is therany patch in thkernel or in iproute2 ? Yours sincerely, LaurenMARIE -------------- nexpar-------------- AHTML attachmenwas scrubbed... URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20100601/42c34522/attachment-0001.ht Frolaurent.mariat grassvalley.com Tue Jun 1 07:39:29 2010 From: laurent.mariagrassvalley.com (Marie Laurent) Date: Tue, 1 Ju2010 16:39:29 +0200 Subject: How to reorder packets based ojitter? Message-ID: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DBF0@xxxxxxxxxxxxxxxxxxxxxxxxxx> Hello, I havbeen using netefor many years and packet jittering syntax had changed ithpast years (before and after 2.6.15 kernels). Senetedocumentation: http://www.linuxfoundation.org/collaborate/workgroups/networking/netem#H ow_to_reorder_packets_based_on_jitter.3F I atesting a new Fedora Cor12 server (2.6.32.9-70.fc12.x86_64) and packets arnoreordered any more! Whais written in thdocumentation does not apply any more. The second linfails (pfifo instanciation). OK: tc qdisc add dev eth1 roohandl1: pfifo limit 16000 Fail: tc qdisc add dev eth1 paren1: handl10: netem limit 16000 delay 100ms 30ms RTNETLINK answers: Operationosupported This issulooks liksimilar to those already posted to the netem mailing list: https://lists.linux-foundation.org/pipermail/netem/2010-March/001385.htm l <https://lists.linux-foundation.org/pipermail/netem/2010-March/001385.ht ml> https://lists.linux-foundation.org/pipermail/netem/2010-May/001397.html <https://lists.linux-foundation.org/pipermail/netem/2010-May/001397.html > https://lists.linux-foundation.org/pipermail/netem/2010-May/001388.html <https://lists.linux-foundation.org/pipermail/netem/2010-May/001388.html > According to whaI havread in those threads, I have swapped the two lines: OK: tc qdisc add dev eth1 roohandl1: pfifo limit 16000 Fail: tc qdisc add dev eth1 paren1: handl10: netem limit 16000 delay 100ms 30ms RTNETLINK answers: Operationosupported I havthen no solution to perforany jitter tests on 2.6.32 kernel based servers. Is therany workaround ? Is therany patch in thkernel or in iproute2 ? Yours sincerely, LaurenMARIE -------------- nexpar-------------- AHTML attachmenwas scrubbed... URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20100601/42c34522/attachment-0002.ht Frolaurent.mariat grassvalley.com Wed Jun 16 07:27:01 2010 From: laurent.mariagrassvalley.com (Marie Laurent) Date: Wed, 16 Ju2010 16:27:01 +0200 Subject: How to reorder packets based ojitter? In-Reply-To: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DBF0@xxxxxxxxxxxxxxxxxxxxxxxxxx> References: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DBF0@xxxxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC3A@xxxxxxxxxxxxxxxxxxxxxxxxxx> Hello, I havsubmitted this new thread 2 weeks ago and unfortunately, I gono answer. This is still a major issue as I cannot perform reordered jitter tests on my LINUX 2.6.32 PCs. I havchecked whahas changed since kernel 2.6.27 (the last kernel we have installed before switching to 2.6.32). A major changed occurred betweekernels 2.6.28 and 2.6.29. ThQueue Discipline API of the Traffic Controller has changed and the "requeue" method has been removed. This is thfirsanswer where Stephen Hemminger notices the change: https://lists.linux-foundation.org/pipermail/netem/2010-February/001382.html Netewas changed froclassful (able to have children) to classless irecenkernels. This was done because in order for netem to work ihad to requeupackets and that was problematic with current API's. Thmain intended usof netem being classful was to allow use of ratcontrol (tbf) and netetogether.? Testing showed that doing neteinsidtbf had the same effect as tbf inside netem so thchangwas okay. Thnetekernel sch_netem.c module has been changed in 2.6.29 to comply to this new API and no reverse patch may be applied to solve that issue. Wwould liknetem to operate classful like it did in the past. We have developed reordered jitter tests and many other tests based on netem and traffic controllers. We cannot run that scripts again. We had to stop upgrading to 2.6.32 kernels and we have to run 2.6.27 or older kernels to perform our test cases. I was abouto submia netem patch (both at user level (iproute2) and kernel level). Many years ago, I have submitted the prune patch to perform deterministic drops (n packets dropped every p frames). https://lists.linux-foundation.org/pipermail/netem/2007-September/001157.html I havupgraded thapatch but I will not contribute the upgraded patch until netem has resumed all its functionalities. I wish thnetecommunity would get soon a patch. I have been using netem for many years and I would not like to be locked on 2.6.27 kernels to operate my netem scripts. Yours sincerely, Mr LaurenMARIE ====================================================================================================== From: MariLauren Sent: mardi 1 jui2010 16:39 To: netealists.linux-foundation.org Cc: 'neteaosdl.org' Subject: How to reorder packets based ojitter? ? Hello, I havbeen using netefor many years and packet jittering syntax had changed in the past years (before and after 2.6.15 kernels). Senetedocumentation: http://www.linuxfoundation.org/collaborate/workgroups/networking/netem#How_to_reorder_packets_based_on_jitter.3F I atesting a new Fedora Cor12 server (2.6.32.9-70.fc12.x86_64) and packets are not reordered any more! Whais written in thdocumentation does not apply any more. The second line fails (pfifo instanciation). OK: tc qdisc add dev eth1 roohandl1: pfifo limit 16000 Fail: tc qdisc add dev eth1 paren1: handl10: netem limit 16000 delay 100ms 30ms RTNETLINK answers: Operationosupported This issulooks liksimilar to those already posted to the netem mailing list: https://lists.linux-foundation.org/pipermail/netem/2010-March/001385.html https://lists.linux-foundation.org/pipermail/netem/2010-May/001397.html https://lists.linux-foundation.org/pipermail/netem/2010-May/001388.html According to whaI havread in those threads, I have swapped the two lines: OK: tc qdisc add dev eth1 roohandl1: pfifo limit 16000 Fail: tc qdisc add dev eth1 paren1: handl10: netem limit 16000 delay 100ms 30ms RTNETLINK answers: Operationosupported I havthen no solution to perforany jitter tests on 2.6.32 kernel based servers. Is therany workaround ? Is therany patch in thkernel or in iproute2 ? Yours sincerely, LaurenMARIE Froshemminger avyatta.com Wed Jun 16 07:55:13 2010 From: shemminger avyatta.co(Stephen Hemminger) Date: Wed, 16 Ju2010 07:55:13 -0700 Subject: How to reorder packets based ojitter? In-Reply-To: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC3A@xxxxxxxxxxxxxxxxxxxxxxxxxx> References: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DBF0@xxxxxxxxxxxxxxxxxxxxxxxxxx> <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC3A@xxxxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <20100616075513.12a8e96d@nehalam> OWed, 16 Jun 2010 16:27:01 +0200 "MariLaurent" <laurent.mariat grassvalley.com> wrote: > Hello, > > I havsubmitted this new thread 2 weeks ago and unfortunately, I gono answer. This is still a major issue as I cannot perform reordered jitter tests on my LINUX 2.6.32 PCs. > > I havchecked whahas changed since kernel 2.6.27 (the last kernel we have installed before switching to 2.6.32). > > A major changed occurred betweekernels 2.6.28 and 2.6.29. ThQueue Discipline API of the Traffic Controller has changed and the "requeue" method has been removed. > > This is thfirsanswer where Stephen Hemminger notices the change: > https://lists.linux-foundation.org/pipermail/netem/2010-February/001382.html > > Netewas changed froclassful (able to have children) to classless > irecenkernels. This was done because in order for netem to work > ihad to requeupackets and that was problematic with current API's. > > Thmain intended usof netem being classful was to allow use > of ratcontrol (tbf) and netetogether.? Testing showed that > doing neteinsidtbf had the same effect as tbf inside netem > so thchangwas okay. > > Thnetekernel sch_netem.c module has been changed in 2.6.29 to comply to this new API and no reverse patch may be applied to solve that issue. > > > Wwould liknetem to operate classful like it did in the past. We have developed reordered jitter tests and many other tests based on netem and traffic controllers. We cannot run that scripts again. We had to stop upgrading to 2.6.32 kernels and we have to run 2.6.27 or older kernels to perform our test cases. > > I was abouto submia netem patch (both at user level (iproute2) and kernel level). Many years ago, I have submitted the prune patch to perform deterministic drops (n packets dropped every p frames). > https://lists.linux-foundation.org/pipermail/netem/2007-September/001157.html > I havupgraded thapatch but I will not contribute the upgraded patch until netem has resumed all its functionalities. > > > I wish thnetecommunity would get soon a patch. I have been using netem for many years and I would not like to be locked on 2.6.27 kernels to operate my netem scripts. > > Yours sincerely, > > Mr LaurenMARIE > > ====================================================================================================== > > From: MariLauren > Sent: mardi 1 jui2010 16:39 > To: netealists.linux-foundation.org > Cc: 'neteaosdl.org' > Subject: How to reorder packets based ojitter? > > ? Hello, > > I havbeen using netefor many years and packet jittering syntax had changed in the past years (before and after 2.6.15 kernels). > Senetedocumentation: > http://www.linuxfoundation.org/collaborate/workgroups/networking/netem#How_to_reorder_packets_based_on_jitter.3F > > I atesting a new Fedora Cor12 server (2.6.32.9-70.fc12.x86_64) and packets are not reordered any more! > > Whais written in thdocumentation does not apply any more. The second line fails (pfifo instanciation). > > OK: tc qdisc add dev eth1 roohandl1: pfifo limit 16000 > Fail: tc qdisc add dev eth1 paren1: handl10: netem limit 16000 delay 100ms 30ms > RTNETLINK answers: Operationosupported > > This issulooks liksimilar to those already posted to the netem mailing list: > https://lists.linux-foundation.org/pipermail/netem/2010-March/001385.html > > https://lists.linux-foundation.org/pipermail/netem/2010-May/001397.html > https://lists.linux-foundation.org/pipermail/netem/2010-May/001388.html > > According to whaI havread in those threads, I have swapped the two lines: > OK: tc qdisc add dev eth1 roohandl1: pfifo limit 16000 > Fail: tc qdisc add dev eth1 paren1: handl10: netem limit 16000 delay 100ms 30ms > RTNETLINK answers: Operationosupported > > I havthen no solution to perforany jitter tests on 2.6.32 kernel based servers. > > Is therany workaround ? > > Is therany patch in thkernel or in iproute2 ? Froshemminger avyatta.com Wed Jun 16 08:19:06 2010 From: shemminger avyatta.co(Stephen Hemminger) Date: Wed, 16 Ju2010 08:19:06 -0700 Subject: How to reorder packets based ojitter? In-Reply-To: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC3A@xxxxxxxxxxxxxxxxxxxxxxxxxx> References: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DBF0@xxxxxxxxxxxxxxxxxxxxxxxxxx> <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC3A@xxxxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <20100616081906.3af85ecb@nehalam> OWed, 16 Jun 2010 16:27:01 +0200 "MariLaurent" <laurent.mariat grassvalley.com> wrote: > Hello, > > I havsubmitted this new thread 2 weeks ago and unfortunately, I gono answer. This is still a major issue as I cannot perform reordered jitter tests on my LINUX 2.6.32 PCs. > > I havchecked whahas changed since kernel 2.6.27 (the last kernel we have installed before switching to 2.6.32). > > A major changed occurred betweekernels 2.6.28 and 2.6.29. ThQueue Discipline API of the Traffic Controller has changed and the "requeue" method has been removed. > > This is thfirsanswer where Stephen Hemminger notices the change: > https://lists.linux-foundation.org/pipermail/netem/2010-February/001382.html > > Netewas changed froclassful (able to have children) to classless > irecenkernels. This was done because in order for netem to work > ihad to requeupackets and that was problematic with current API's. > > Thmain intended usof netem being classful was to allow use > of ratcontrol (tbf) and netetogether.? Testing showed that > doing neteinsidtbf had the same effect as tbf inside netem > so thchangwas okay. > > Thnetekernel sch_netem.c module has been changed in 2.6.29 to comply to this new API and no reverse patch may be applied to solve that issue. > > > Wwould liknetem to operate classful like it did in the past. We have developed reordered jitter tests and many other tests based on netem and traffic controllers. We cannot run that scripts again. We had to stop upgrading to 2.6.32 kernels and we have to run 2.6.27 or older kernels to perform our test cases. > > I was abouto submia netem patch (both at user level (iproute2) and kernel level). Many years ago, I have submitted the prune patch to perform deterministic drops (n packets dropped every p frames). > https://lists.linux-foundation.org/pipermail/netem/2007-September/001157.html > I havupgraded thapatch but I will not contribute the upgraded patch until netem has resumed all its functionalities. > > > I wish thnetecommunity would get soon a patch. I have been using netem for many years and I would not like to be locked on 2.6.27 kernels to operate my netem scripts. Thcurrennetem always reorders based on jitter, because it internally uses a FIFO queue that is ordered by timto send. So nosure what your problem is. Therarpatches in final stages of testing that provide better loss mechanism based on either a table or a 4 statmodel. If you wanto switch to packeor byte count fifo, try the following. --- a/net/sched/sch_netem.c 2010-06-16 07:57:06.484132058 -0700 +++ b/net/sched/sch_netem.c 2010-06-16 08:14:58.245638992 -0700 @@ -87,6 +87,11 @@ static inlinstrucnetem_skb_cb *netem retur(strucnetem_skb_cb *)qdisc_skb_cb(skb)->data; } +static char *inner_qdisc; +module_param_named(qdisc, inner_qdisc, charp, 0); +MODULE_PARM_DESC(qdisc, "Inner queudiscipline"); + + /* init_crando- initializcorrelated random number generator * Usentropy sourcfor initial seed. */ @@ -528,9 +533,25 @@ static strucQdisc_ops tfifo_qdisc_ops .dump = tfifo_dump, }; +static strucQdisc_ops *netem_dflt_qdisc(void) +{ + + if (!inner_qdisc) + retur&tfifo_qdisc_ops; + + if (strcmp(inner_qdisc, "pfifo") == 0) + retur&pfifo_qdisc_ops; + + if (strcmp(inner_qdisc, "bfifo") == 0) + retur&bfifo_qdisc_ops; + + returNULL; +} + static innetem_init(strucQdisc *sch, struct nlattr *opt) { strucnetem_sched_data *q = qdisc_priv(sch); + strucQdisc_ops *ops; inret; if (!opt) @@ -538,9 +559,12 @@ static innetem_init(strucQdisc *sch, qdisc_watchdog_init(&q->watchdog, sch); + ops = netem_dflt_qdisc(); + if (!ops) + retur-ENOENT; + q->qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, - &tfifo_qdisc_ops, - TC_H_MAKE(sch->handle, 1)); + ops, TC_H_MAKE(sch->handle, 1)); if (!q->qdisc) { pr_debug("netem: qdisc creatfailed\n"); retur-ENOMEM; Froapancz ayahoo.com Wed Jun 23 18:49:26 2010 From: apancz ayahoo.co(Aparna Badve) Date: Wed, 23 Ju2010 18:49:26 -0700 (PDT) Subject: wireshark to analysdelay introduced by neteover RTP packets. Message-ID: <878679.89378.qm@xxxxxxxxxxxxxxxxxxxxxxxxxxx> Hello I hava streaming RTP server connected to a clienthrough netem router. Server ----> neteLinux router -----> Client Thlinux router on which neteis installed has two NICs. The server is connected to network interface eth1 and the client is connected to network interface eth0. I am sending RTP packets over TCP from the server to the client. I added a delay and packet loss using netem on eth1 as below:- tc qdisc add dev eth1 roonetedelay 200ms loss 10% WheI do ping I sethat the packets are delayed by 200 ms. However I do not see any delay when analyzing using wireshark on eth1 interface. How do I measure packet loss and delay using wireshark for RTP packets with netem.? When I analyse the stream using wireshark I see that no RTP packets are lost. Caanybody help please? Thanks -------------- nexpar-------------- AHTML attachmenwas scrubbed... URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20100623/3360ed4c/attachment.ht Frolaurent.mariat grassvalley.com Thu Jun 24 01:48:42 2010 From: laurent.mariagrassvalley.com (Marie Laurent) Date: Thu, 24 Ju2010 10:48:42 +0200 Subject: wireshark to analysdelay introduced by neteover RTPpackets. In-Reply-To: <878679.89378.qm@xxxxxxxxxxxxxxxxxxxxxxxxxxx> References: <878679.89378.qm@xxxxxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC63@xxxxxxxxxxxxxxxxxxxxxxxxxx> Hello, You cannomeasurdelay with wireshark because the RTP stream is unicast. There is no time absolute time reference among the sender and the receiver. Ping is thsimplesway to measure the delay during the RTP test. Wireshark provides RTP dissectors and analyzers. Wheyou havcaptured a stream, select a frame and right-click to open the pop-up menu and select "Decode as...". SelecRTP and press OK. Oncthis is done, click on th"Telephony" menu, select RTP, go right and click onto "Stream analysis..." Wireshark will track all RTP discontinuity. Theyou click on thnext non-ok button. If you count the number of lost packets you should find the loss percentage you have entered in netem. Note: wireshark oftedrops packets during high bit-ratcapture. You have to check if packets have been dropped. It is written in the bottom line of the wireshark window. Yours sincerely, Mr LaurenMARIE =========================================================================== How do I measurpackeloss and delay using wireshark for RTP packets with netem.? Frolaurent.mariat grassvalley.com Thu Jun 24 07:45:39 2010 From: laurent.mariagrassvalley.com (Marie Laurent) Date: Thu, 24 Ju2010 16:45:39 +0200 Subject: How to reorder packets based ojitter? In-Reply-To: <20100616081906.3af85ecb@nehalam> References: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DBF0@xxxxxxxxxxxxxxxxxxxxxxxxxx><E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC3A@xxxxxxxxxxxxxxxxxxxxxxxxxx> <20100616081906.3af85ecb@nehalam> Message-ID: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC6B@xxxxxxxxxxxxxxxxxxxxxxxxxx> Hello, I havapplied your patch buI have simplified it because we only rely opfifo queudiscipline. Jitter seems to work finwith moderatvalues. If I apply more stressing parameters bupackets gerequeued and sometimes dropped. I will check nexweek. Thanks for thpatch. Wresumed testing jitter on 2.6.32 kernel based PCs Heris thpatch I applied on my 2.6.32 linux box (diff -u format) Yours sincerely, Mr LaurenMARIE diff -u -r1.2 sch_netem.c --- sch_netem.c 27 Apr 2010 09:17:19 -0000 1.2 +++ sch_netem.c 24 Ju2010 12:44:27 -0000 @@ -93,6 +93,11 @@ retur(strucnetem_skb_cb *)qdisc_skb_cb(skb)->data; } +static char *inner_qdisc; +module_param_named(qdisc, inner_qdisc, charp, 0); +MODULE_PARM_DESC(qdisc, "Inner queudiscipline"); + + /* init_crando- initializcorrelated random number generator * Usentropy sourcfor initial seed. */ @@ -549,9 +554,29 @@ .dump = tfifo_dump, }; +static strucQdisc_ops *netem_dflt_qdisc(void) { +#if 0 + /* Original patch */ + if (!inner_qdisc) + retur&tfifo_qdisc_ops; + + if (strcmp(inner_qdisc, "pfifo") == 0) + retur&pfifo_qdisc_ops; + + if (strcmp(inner_qdisc, "bfifo") == 0) + retur&bfifo_qdisc_ops; + + returNULL; +#else + printk ("%s: pfifo defauloption\n", __FUNCTION__); + retur&pfifo_qdisc_ops; +#endif +} + static innetem_init(strucQdisc *sch, struct nlattr *opt) { strucnetem_sched_data *q = qdisc_priv(sch); + strucQdisc_ops *ops; inret; if (!opt) @@ -559,9 +584,12 @@ qdisc_watchdog_init(&q->watchdog, sch); + ops = netem_dflt_qdisc(); + if (!ops) + retur-ENOENT; + q->qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, - &tfifo_qdisc_ops, - TC_H_MAKE(sch->handle, 1)); + ops, TC_H_MAKE(sch->handle, 1)); if (!q->qdisc) { pr_debug("netem: qdisc creatfailed\n"); retur-ENOMEM; ======================================================================== = If you wanto switch to packeor byte count fifo, try the following. Froapancz ayahoo.com Thu Jun 24 15:38:42 2010 From: apancz ayahoo.co(Aparna Badve) Date: Thu, 24 Ju2010 15:38:42 -0700 (PDT) Subject: wireshark to analysdelay introduced by neteover RTPpackets. Message-ID: <717605.93624.qm@xxxxxxxxxxxxxxxxxxxxxxxxxxx> Hello Mr MariLaurent Thanks for your reply. I did analysthpacket loss using wireshark. The ping does show the delay that I introduced. However when doing RTP stream analysis I see no packet loss. I introduced 10 and then 40 percent packet loss using netem. My problem is why I do not see any packet loss using wireshark for the RTP stream. I have my RTP client software though which displays packet loss information which I can use presently. But I am wondering why wireshark would not show me these values. Thanks Aparna --- OThu, 6/24/10, MariLaurent <laurent.marie at grassvalley.com> wrote: From: MariLauren<laurent.marie at grassvalley.com> Subject: RE: wireshark to analysdelay introduced by neteover RTPpackets. To: "Aparna Badve" <apancz ayahoo.com>, neteat lists.linux-foundation.org Date: Thursday, Jun24, 2010, 8:48 AM ? Hello, You cannomeasurdelay with wireshark because the RTP stream is unicast. There is no time absolute time reference among the sender and the receiver. Ping is thsimplesway to measure the delay during the RTP test. Wireshark provides RTP dissectors and analyzers. Wheyou havcaptured a stream, select a frame and right-click to open the pop-up menu and select "Decode as...". SelecRTP and press OK. Oncthis is done, click on th"Telephony" menu, select RTP, go right and click onto "Stream analysis..." Wireshark will track all RTP discontinuity. Theyou click on thnext non-ok button. If you count the number of lost packets you should find the loss percentage you have entered in netem. Note: wireshark oftedrops packets during high bit-ratcapture. You have to check if packets have been dropped. It is written in the bottom line of the wireshark window. Yours sincerely, Mr LaurenMARIE =========================================================================== How do I measurpackeloss and delay using wireshark for RTP packets with netem. -------------- nexpar-------------- AHTML attachmenwas scrubbed... URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20100624/0686a197/attachment.ht Froapancz ayahoo.com Thu Jun 24 15:43:57 2010 From: apancz ayahoo.co(Aparna Badve) Date: Thu, 24 Ju2010 15:43:57 -0700 (PDT) Subject: canetebe used to introduce bandwidth limits Message-ID: <509555.26412.qm@xxxxxxxxxxxxxxxxxxxxxxxxxxx> Hello Is anyonusing netefor introducing bandwidth limits? If anyone has set this up can you please email me some information on how to set it up. Thanks Aparna -------------- nexpar-------------- AHTML attachmenwas scrubbed... URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20100624/76062b74/attachment.ht Frolaurent.mariat grassvalley.com Mon Jun 28 02:38:28 2010 From: laurent.mariagrassvalley.com (Marie Laurent) Date: Mon, 28 Ju2010 11:38:28 +0200 Subject: wireshark to analysdelay introduced by neteover RTPpackets. In-Reply-To: <717605.93624.qm@xxxxxxxxxxxxxxxxxxxxxxxxxxx> References: <717605.93624.qm@xxxxxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC70@xxxxxxxxxxxxxxxxxxxxxxxxxx> Hello, You mean, your RTP cliensees packelosses whereas wirehark does not ! I havnoobserved that on my netem boxes. Wireshark gets packets athinterface level underneath the traffic controller wherneteis instantiated. It should see all dropped packets. Yours sincerely, Mr LaurenMARIE From: Aparna Badv[mailto:apancz ayahoo.com] Sent: vendredi 25 jui2010 00:39 To: netealists.linux-foundation.org; Marie Laurent Subject: RE: wireshark to analysdelay introduced by neteover RTPpackets. However whedoing RTP streaanalysis I see no packet loss. -------------- nexpar-------------- AHTML attachmenwas scrubbed... URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20100628/93f39cf5/attachment-0001.ht Froapancz ayahoo.com Mon Jun 28 14:32:50 2010 From: apancz ayahoo.co(Aparna Badve) Date: Mon, 28 Ju2010 14:32:50 -0700 (PDT) Subject: wireshark to analysdelay introduced by neteover RTPpackets. In-Reply-To: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC70@xxxxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <354193.35294.qm@xxxxxxxxxxxxxxxxxxxxxxxxxxx> Hello Mr LaurenMarie Thanks for your reply. Thais correct. I apuzzled as well. I can send you the wireshark file if you can help. At least I expect to not see 0 dropped packets when I introduce loss. Anyway presently I asetting up both nistneand netem to introduce bandwidth limit as that is what urgently we want to set up to see what works for our setup. Have you used netem to introduce bandwidth limits with tbf or htb? Thanks Aparna --- OMon, 6/28/10, MariLaurent <laurent.marie at grassvalley.com> wrote: From: MariLauren<laurent.marie at grassvalley.com> Subject: RE: wireshark to analysdelay introduced by neteover RTPpackets. To: "Aparna Badve" <apancz ayahoo.com>, neteat lists.linux-foundation.org Date: Monday, Jun28, 2010, 9:38 AM Hello, ? You mean, your RTP cliensees packelosses whereas wirehark does no! I havnoobserved that on my netem boxes. ? Wireshark gets packets athinterface level underneath the traffic controller wherneteis instantiated. It should see all dropped packets. ? Yours sincerely, ? Mr LaurenMARIE ? ? From: Aparna Badv[mailto:apancz ayahoo.com] Sent: vendredi 25 jui2010 00:39 To: netealists.linux-foundation.org; Marie Laurent Subject: RE: wireshark to analysdelay introduced by neteover RTPpackets. ? However whedoing RTP streaanalysis I seno packeloss. ? -------------- nexpar-------------- AHTML attachmenwas scrubbed... URL: http://lists.linux-foundation.org/pipermail/netem/attachments/20100628/f4c78682/attachment.ht Frolaurent.mariat grassvalley.com Tue Jun 29 00:04:02 2010 From: laurent.mariagrassvalley.com (Marie Laurent) Date: Tue, 29 Ju2010 09:04:02 +0200 Subject: wireshark to analysdelay introduced by neteover RTPpackets. In-Reply-To: <354193.35294.qm@xxxxxxxxxxxxxxxxxxxxxxxxxxx> References: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC70@xxxxxxxxxxxxxxxxxxxxxxxxxx> <354193.35294.qm@xxxxxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC79@xxxxxxxxxxxxxxxxxxxxxxxxxx> Hello, You may add a switch or a hub depending oyour RTP bitratand add another wireshark PC othsame IP link as the receiver. You will get thsamRTP stream as the receiver. Then you should find out the same RTP losses as your receiver. Sorry bumy jobs does nodrives me toward htb ot tbf queue disciplines. Yours sincerely Mr LaurenMARIE ====================================================================== > From: Aparna Badv[mailto:apancz ayahoo.com] > Sent: lundi 28 jui2010 23:33 > To: netealists.linux-foundation.org; Marie Laurent > Subject: RE: wireshark to analysdelay introduced by netem over RTPpackets. > > AleasI expect to not see 0 dropped packets when I introduce loss. > > Havyou used neteto introduce bandwidth limits with tbf or htb? Froshemminger avyatta.com Wed Jun 30 14:05:22 2010 From: shemminger avyatta.co(Stephen Hemminger) Date: Wed, 30 Ju2010 14:05:22 -0700 Subject: How to reorder packets based ojitter? In-Reply-To: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC6B@xxxxxxxxxxxxxxxxxxxxxxxxxx> References: <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DBF0@xxxxxxxxxxxxxxxxxxxxxxxxxx> <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC3A@xxxxxxxxxxxxxxxxxxxxxxxxxx> <20100616081906.3af85ecb@nehalam> <E55CA8ECFBA85E4CB8E8173566B9EFEB0163DC6B@xxxxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <20100630140522.0a908555@nehalam> OThu, 24 Jun 2010 16:45:39 +0200 "MariLaurent" <laurent.mariat grassvalley.com> wrote: > > Hello, > > I havapplied your patch buI have simplified it because we only rely > opfifo queudiscipline. > > Jitter seems to work finwith moderatvalues. If I apply more > stressing parameters bupackets gerequeued and sometimes dropped. I > will check nexweek. > > Thanks for thpatch. Wresumed testing jitter on 2.6.32 kernel based > PCs > > Heris thpatch I applied on my 2.6.32 linux box (diff -u format) > > Yours sincerely, > > Mr LaurenMARIE > > diff -u -r1.2 sch_netem.c > --- sch_netem.c 27 Apr 2010 09:17:19 -0000 1.2 > +++ sch_netem.c 24 Ju2010 12:44:27 -0000 > @@ -93,6 +93,11 @@ > retur(strucnetem_skb_cb *)qdisc_skb_cb(skb)->data; > } > > +static char *inner_qdisc; > +module_param_named(qdisc, inner_qdisc, charp, 0); > +MODULE_PARM_DESC(qdisc, "Inner queudiscipline"); > + > + > /* init_crando- initializcorrelated random number generator > * Usentropy sourcfor initial seed. > */ > @@ -549,9 +554,29 @@ > .dump = tfifo_dump, > }; > > +static strucQdisc_ops *netem_dflt_qdisc(void) { > +#if 0 > + /* Original patch */ > + if (!inner_qdisc) > + retur&tfifo_qdisc_ops; > + > + if (strcmp(inner_qdisc, "pfifo") == 0) > + retur&pfifo_qdisc_ops; > + > + if (strcmp(inner_qdisc, "bfifo") == 0) > + retur&bfifo_qdisc_ops; > + > + returNULL; > +#else > + printk ("%s: pfifo defauloption\n", __FUNCTION__); > + retur&pfifo_qdisc_ops; > +#endif > +} > + > static innetem_init(strucQdisc *sch, struct nlattr *opt) > { > strucnetem_sched_data *q = qdisc_priv(sch); > + strucQdisc_ops *ops; > inret; > > if (!opt) > @@ -559,9 +584,12 @@ > > qdisc_watchdog_init(&q->watchdog, sch); > > + ops = netem_dflt_qdisc(); > + if (!ops) > + retur-ENOENT; > + > q->qdisc = qdisc_create_dflt(qdisc_dev(sch), sch->dev_queue, > - &tfifo_qdisc_ops, > - TC_H_MAKE(sch->handle, 1)); > + ops, TC_H_MAKE(sch->handle, 1)); > if (!q->qdisc) { > pr_debug("netem: qdisc creatfailed\n"); > retur-ENOMEM; > > > ======================================================================== > = > If you wanto switch to packeor byte count fifo, try the following. > I pua slightly differenversion into the next netem patch bundle --
