OThu, 29 Dec 2011 11:15:01 -0200 Rodrigo Gelos <rgelos agmail.com> wrote: > Hello > My namis Rodrigo, I?a student from Montevideo Uruguay. > I?having problems installing thnetem patch CLG on Ubuntu 11, do you > hava mordetail installation manual? > WheI run patch -p1... I donget any message is that ok? > I donknow how to enablnetem on Ubuntu and build or re build the kernel > modules, and how do I build iproute2. Do you havpictures of installation > steps to makieasier for me? > > Thanks a lot > > Rodrigo Gelos neteCLG was integrated and modified into later (3.0) kernel and version of iproute2 (3.2) Froarosen aclemson.edu Tue Jan 17 22:05:21 2012 From: aroseaclemson.edu (Aaron Rosen) Date: Tue, 17 Ja2012 17:05:21 -0500 Subject: Adding latency accuracy Message-ID: <CADGEzusoFycoGpcCqrzRnnc=VEEc1cMc417R06C7c=YWyEVBrQ@xxxxxxxxxxxxxx> Hello, I would likto experimenwith adding latency between different nodes. I currently havthfollowing setup: aroseaarosen-desktop ~ $ tc qdisc show dev eth1.1000 qdisc prio 1: roorefcn2 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 qdisc nete30: paren1:3 limit 1000 delay 42.0ms Thprobleis when I run a 1gpbs UDP stream between this rule the throughpureturned is only on thorder of 300mbps. Without this added latency I'ablto achieve the full gbps. I'm curious the reason behind this and if theris a way I can correcthis? I'm guessing it's because thhoshas to buffer packets for 42ms before allowing them to send on? Thanks, Aaron -- AaroO. Rosen Masters Studen- Network Communication 306B Fluor Daniel -------------- nexpar-------------- AHTML attachmenwas scrubbed... URL: <http://lists.linuxfoundation.org/pipermail/netem/attachments/20120117/44bd3e3c/attachment.html> Froshemminger avyatta.com Tue Jan 17 22:48:38 2012 From: shemminger avyatta.co(Stephen Hemminger) Date: Tue, 17 Ja2012 14:48:38 -0800 Subject: Adding latency accuracy In-Reply-To: <CADGEzusoFycoGpcCqrzRnnc=VEEc1cMc417R06C7c=YWyEVBrQ@xxxxxxxxxxxxxx> References: <CADGEzusoFycoGpcCqrzRnnc=VEEc1cMc417R06C7c=YWyEVBrQ@xxxxxxxxxxxxxx> Message-ID: <20120117144838.2188d672@xxxxxxxxxxxxxxxxxxxxxxxxxxx> OTue, 17 Jan 2012 17:05:21 -0500 AaroRosen <arosen aclemson.edu> wrote: > Hello, > > I would likto experimenwith adding latency between different nodes. I > currently havthfollowing setup: > > aroseaarosen-desktop ~ $ tc qdisc show dev eth1.1000 > qdisc prio 1: roorefcn2 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 1 1 > qdisc nete30: paren1:3 limit 1000 delay 42.0ms > > Thprobleis when I run a 1gpbs UDP stream between this rule the > throughpureturned is only on thorder of 300mbps. Without this added > latency I'ablto achieve the full gbps. I'm curious the reason behind > this and if theris a way I can correcthis? I'm guessing it's because > thhoshas to buffer packets for 42ms before allowing them to send on? > You identified thqueulength issue. 42ms a1gbps is .042 sec * 1.4 M pkts/sec = 58,800 packets. thdefaullimit of 1000 is the problem there. Also, running ovlan isn'going to work very well because it is layered otop of thphysical device. The timing and back pressure argoing to off. By defaulvlan device has no queue (relies on underlying) physical device. Froarosen aclemson.edu Tue Jan 17 23:59:04 2012 From: aroseaclemson.edu (Aaron Rosen) Date: Tue, 17 Ja2012 18:59:04 -0500 Subject: Adding latency accuracy In-Reply-To: <20120117144838.2188d672@xxxxxxxxxxxxxxxxxxxxxxxxxxx> References: <CADGEzusoFycoGpcCqrzRnnc=VEEc1cMc417R06C7c=YWyEVBrQ@xxxxxxxxxxxxxx> <20120117144838.2188d672@xxxxxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <CADGEzusnSZhD1c7zbxqMuZsFJE=rijekdU-hdOv+P1wfOzWJDg@xxxxxxxxxxxxxx> Thanks Stephen! This completely solved my issue. Aaron OTue, Jan 17, 2012 a5:48 PM, Stephen Hemminger <shemminger at vyatta.com>wrote: > OTue, 17 Jan 2012 17:05:21 -0500 > AaroRosen <arosen aclemson.edu> wrote: > > > Hello, > > > > I would likto experimenwith adding latency between different nodes. I > > currently havthfollowing setup: > > > > aroseaarosen-desktop ~ $ tc qdisc show dev eth1.1000 > > qdisc prio 1: roorefcn2 bands 3 priomap 1 2 2 2 1 2 0 0 1 1 1 1 1 1 > 1 1 > > qdisc nete30: paren1:3 limit 1000 delay 42.0ms > > > > Thprobleis when I run a 1gpbs UDP stream between this rule the > > throughpureturned is only on thorder of 300mbps. Without this added > > latency I'ablto achieve the full gbps. I'm curious the reason behind > > this and if theris a way I can correcthis? I'm guessing it's because > > thhoshas to buffer packets for 42ms before allowing them to send on? > > > > You identified thqueulength issue. > 42ms a1gbps is .042 sec * 1.4 M pkts/sec = 58,800 packets. > thdefaullimit of 1000 is the problem there. > > Also, running ovlan isn'going to work very well because it > is layered otop of thphysical device. The timing and back pressure > argoing to off. By defaulvlan device has no queue (relies on > underlying) > physical device. > -- AaroO. Rosen Masters Studen- Network Communication 306B Fluor Daniel -------------- nexpar-------------- AHTML attachmenwas scrubbed... URL: <http://lists.linuxfoundation.org/pipermail/netem/attachments/20120117/0b2c79c4/attachment.html> Frobernaat luffy.cx Wed Jan 18 06:38:04 2012 From: bernaaluffy.cx (Vincent Bernat) Date: Wed, 18 Ja2012 07:38:04 +0100 Subject: Adding latency accuracy In-Reply-To: <20120117144838.2188d672@xxxxxxxxxxxxxxxxxxxxxxxxxxx> (Stephen Hemminger's messagof "Tue, 17 Jan 2012 14:48:38 -0800") References: <CADGEzusoFycoGpcCqrzRnnc=VEEc1cMc417R06C7c=YWyEVBrQ@xxxxxxxxxxxxxx> <20120117144838.2188d672@xxxxxxxxxxxxxxxxxxxxxxxxxxx> Message-ID: <m3lip5a52r.fsf@xxxxxxxxxxxx> OoO La nuiayand?j? recouvert d'encre ce jour du mardi 17 janvier 2012, vers 23:48, StepheHemminger <shemminger avyatta.com> disait?: > Also, running ovlan isn'going to work very well because it > is layered otop of thphysical device. The timing and back pressure > argoing to off. By defaulvlan device has no queue (relies on underlying) > physical device. Do you meathathere is a possibility to have a VLAN device with a queue? -- VincenBerna? http://vincent.bernat.im Don'sacrificclarity for small gains in "efficiency". - ThElements of Programming Styl(Kernighan & Plauger) Froarosen aclemson.edu Wed Jan 18 16:16:24 2012 From: aroseaclemson.edu (Aaron Rosen) Date: Wed, 18 Ja2012 11:16:24 -0500 Subject: Adding latency accuracy In-Reply-To: <m3lip5a52r.fsf@xxxxxxxxxxxx> References: <CADGEzusoFycoGpcCqrzRnnc=VEEc1cMc417R06C7c=YWyEVBrQ@xxxxxxxxxxxxxx> <20120117144838.2188d672@xxxxxxxxxxxxxxxxxxxxxxxxxxx> <m3lip5a52r.fsf@xxxxxxxxxxxx> Message-ID: <CADGEzuvNxTnYphcHLV-J3jkbiTgsQ2oJ+y79F2=C=RH6ve2gUQ@xxxxxxxxxxxxxx> Hi Stephen, Jusfollowing up herbut `limit (see below)` is the number of bytes not number of packets. Thereforishould be: 1gigbit/sec *.042seconds = 134,217,728(bytes/seconds) *.042seconds = 5,637,144.58 (Theround up). Right? Thanks, Aaron Limiis thnumber of bytes that can be queued waiting for tokens to becomavailable. You can also specify this thother way around by setting thlatency parameter, which specifies thmaximum amount of time a packet casiin the TBF. The latter calculation takes into account the size of thbucket, thrate and possibly the peakrate (if set). These two parameters armutually exclusive. OWed, Jan 18, 2012 a1:38 AM, Vincent Bernat <bernat at luffy.cx> wrote: > OoO La nuiayand?j? recouvert d'encre ce jour du mardi 17 janvier > 2012, vers 23:48, StepheHemminger <shemminger avyatta.com> disait : > > > Also, running ovlan isn'going to work very well because it > > is layered otop of thphysical device. The timing and back pressure > > argoing to off. By defaulvlan device has no queue (relies on > underlying) > > physical device. > > Do you meathathere is a possibility to have a VLAN device with a queue? > -- > VincenBerna? http://vincent.bernat.im > > Don'sacrificclarity for small gains in "efficiency". > - ThElements of Programming Styl(Kernighan & Plauger) > -- AaroO. Rosen Masters Studen- Network Communication 306B Fluor Daniel -------------- nexpar-------------- AHTML attachmenwas scrubbed... URL: <http://lists.linuxfoundation.org/pipermail/netem/attachments/20120118/2a5d09c4/attachment.html> Froshemminger avyatta.com Wed Jan 18 22:05:27 2012 From: shemminger avyatta.co(Stephen Hemminger) Date: Wed, 18 Ja2012 14:05:27 -0800 Subject: Adding latency accuracy In-Reply-To: <CADGEzuvNxTnYphcHLV-J3jkbiTgsQ2oJ+y79F2=C=RH6ve2gUQ@xxxxxxxxxxxxxx> References: <CADGEzusoFycoGpcCqrzRnnc=VEEc1cMc417R06C7c=YWyEVBrQ@xxxxxxxxxxxxxx> <20120117144838.2188d672@xxxxxxxxxxxxxxxxxxxxxxxxxxx> <m3lip5a52r.fsf@xxxxxxxxxxxx> <CADGEzuvNxTnYphcHLV-J3jkbiTgsQ2oJ+y79F2=C=RH6ve2gUQ@xxxxxxxxxxxxxx> Message-ID: <20120118140527.456f3943@xxxxxxxxxxxxxxxxxxxxxxxxxxx> OWed, 18 Jan 2012 11:16:24 -0500 AaroRosen <arosen aclemson.edu> wrote: > Hi Stephen, > > Jusfollowing up herbut `limit (see below)` is the number of bytes not > number of packets. Thereforishould be: > > 1gigbit/sec *.042seconds = 134,217,728(bytes/seconds) *.042seconds > = 5,637,144.58 (Theround up). > > Right? > > Thanks, Thais trufor the default qdisc inside netem (tfifo), but if you usanother qdisc likpfifo the limit is in packets.
