From: Mohammad H. <ha...@pu...> - 2013-10-14 11:20:03
|
>From what I remember, iperf reports the bandwidth from each thread separately but then combines all values into one value at the end of each time-window. I think the combined value from all threads has "-1" in it (if you enable csv); in that case, just grep for ",-1,". -Mohammad On Mon, Oct 14, 2013 at 5:43 AM, Upendra K <upe...@gm...>wrote: > Hi Team > > I have one question , > > !.>when i ran the iperf with 128 thread i am not getting the SUM value , > > example :- > server :- iperf -s -i 1 -B xx.xx.xx.xx |grep SUM > > client :- iperf -c xx.xx.xx. -B xx.xx.xx.xx -i 1 -P 128 |grep SUM > > But if remove the grep SUM , traffic goes fine > > Please find the below details :- > > iperf version :- 2.0.5 > > Please let me know what is the issues , is this issues with the iperf.? > > Regards > Upendra > > > > > On Mon, Oct 14, 2013 at 6:19 AM, Mohammad Hajjat <ha...@pu...>wrote: > >> Yeah I agree, it's not very clear what's going on! But don't you think >> the 160 ms is a bit too high? Maybe the RTT you measure is different for >> ping's ICMP than in TCP (which iperf uses by default)? >> >> >> On Sun, Oct 13, 2013 at 8:13 PM, Nik Koracek <nko...@tp...> wrote: >> >>> ** ** ** >>> >>> Hi Mohammad,**** >>> >>> ** ** >>> >>> I can and I know that I am able to max out the link when I increase the >>> window size and number of parallel streams. My test was focused on running >>> a single tcp stream.**** >>> >>> ** ** >>> >>> My only concern is that I don’t understand the logic behind getting the >>> results that I did. I would have expected to get 1.265Mbps with the test >>> setup the way I described.**** >>> >>> ** ** >>> >>> I would like to understand how I was getting 65Mbps+ on that 160ms RTT >>> link with those default window sizes.**** >>> >>> ** ** >>> >>> Cheers,**** >>> >>> ** ** >>> ------------------------------ >>> >>> *From:* Mohammad Hajjat [mailto:ha...@pu...] >>> *Sent:* Monday, 14 October 2013 11:00 AM >>> *To:* Nik Koracek >>> *Cc:* Mohammad Hajjat; iperf-users >>> >>> *Subject:* Re: [Iperf-users] Help understanding results from a single >>> TCP session >>> **** >>> >>> ** ** >>> >>> I see, why can't you just increase iperf's window size? **** >>> >>> I had experiments where I had to set the window size in iperf to 16 MB >>> to get an optimal bandwidth. **** >>> >>> ** ** >>> >>> On Sun, Oct 13, 2013 at 7:08 PM, Nik Koracek <nko...@tp...> >>> wrote:**** >>> >>> Hi Mohammad,**** >>> >>> **** >>> >>> Thanks for the reply, sorry for any confusion.**** >>> >>> **** >>> >>> During testing the client machine was using a default window size of >>> 23.2 Kbytes, while the server machine was using a default window size of >>> 85.3 Kbytes.**** >>> >>> **** >>> >>> My question is, do you know how I could get in excess of 65Mbps on a >>> single tcp session using the aforementioned window sizes between two >>> devices that had a 160ms RTT between them?**** >>> >>> **** >>> >>> Cheers**** >>> >>> **** >>> ------------------------------ >>> >>> *From:* Mohammad Hajjat [mailto:ha...@pu...] >>> *Sent:* Monday, 14 October 2013 12:18 AM >>> *To:* nko...@tp... >>> *Cc:* iperf-users >>> *Subject:* Re: [Iperf-users] Help understanding results from a single >>> TCP session**** >>> >>> **** >>> >>> But your server's window size isn't 25 KB, is it?**** >>> >>> Also, what is it in specific that you want to explain? I'm sorry I lost >>> track :P**** >>> >>> **** >>> >>> **** >>> >>> On Sun, Oct 13, 2013 at 7:46 AM, <nko...@tp...> wrote:**** >>> >>> Hi All,**** >>> >>> **** >>> >>> I was troubleshooting a throughput issue recently on an international >>> 300Mbps link which had a RTT of ~160ms. **** >>> >>> **** >>> >>> I have copied the output below from testing a single TCP session between >>> two linux servers at either endpoint.**** >>> >>> **** >>> >>> I have the output from two tests from the perspective of SERVER_A >>> (acting first in server mode and then in client mode). The remote server >>> had similar output and same window sizes i.e. when in server mode it >>> defaulted to 85.3 Kbytes and when in client mode defaulted to 25.3 Kbytes. >>> **** >>> >>> **** >>> >>> ******************************************************** >>> >>> SERVER MODE (receiving connection from remote server)**** >>> >>> ******************************************************** >>> >>> **** >>> >>> [root@SERVER_A ~]# iperf -s -i 1**** >>> >>> ------------------------------------------------------------**** >>> >>> Server listening on TCP port 5001**** >>> >>> TCP window size: 85.3 KByte (default)**** >>> >>> ------------------------------------------------------------**** >>> >>> [ 4] local 192.168.1.1 port 5001 connected with 192.168.2.1 port 46326* >>> *** >>> >>> [ ID] Interval Transfer Bandwidth**** >>> >>> [ 4] 0.0- 1.0 sec 491 KBytes 4.02 Mbits/sec**** >>> >>> [ 4] 1.0- 2.0 sec 3.79 MBytes 31.8 Mbits/sec**** >>> >>> [ 4] 2.0- 3.0 sec 9.13 MBytes 76.6 Mbits/sec**** >>> >>> [ 4] 3.0- 4.0 sec 9.01 MBytes 75.6 Mbits/sec**** >>> >>> [ 4] 4.0- 5.0 sec 9.01 MBytes 75.6 Mbits/sec**** >>> >>> [ 4] 5.0- 6.0 sec 10.1 MBytes 84.6 Mbits/sec**** >>> >>> [ 4] 6.0- 7.0 sec 9.00 MBytes 75.5 Mbits/sec**** >>> >>> [ 4] 7.0- 8.0 sec 9.13 MBytes 76.6 Mbits/sec**** >>> >>> [ 4] 8.0- 9.0 sec 9.01 MBytes 75.6 Mbits/sec**** >>> >>> [ 4] 9.0-10.0 sec 9.02 MBytes 75.7 Mbits/sec**** >>> >>> [ 4] 0.0-10.2 sec 79.8 MBytes 65.6 Mbits/sec**** >>> >>> **** >>> >>> ******************************************************** >>> >>> CLIENT MODE (sending data to remote server)**** >>> >>> ******************************************************** >>> >>> **** >>> >>> [root@SERVER_A~]iperf -c 192.168.2.1**** >>> >>> ------------------------------------------------------------**** >>> >>> Client connecting to 192.168.2.1, TCP port 5001**** >>> >>> TCP window size: 23.2 KByte (default)**** >>> >>> ------------------------------------------------------------**** >>> >>> [ 3] local 192.168.1.1 port 56806 connected with 192.168.2.1 port 5001* >>> *** >>> >>> [ ID] Interval Transfer Bandwidth**** >>> >>> [ 3] 0.0-10.1 sec 116 MBytes 96.7 Mbits/sec**** >>> >>> **** >>> >>> **** >>> >>> What I did not understand is how we were getting in excess of 65 Mbps >>> using a single TCP session and default window size of 25.3 Kbytes.**** >>> >>> **** >>> >>> From my calculations **** >>> >>> **** >>> >>> 1000ms/160ms = 6.25 window periods**** >>> >>> 6.25 x window size (25.3Kbytes) = 158.125 Kbytes**** >>> >>> multiplied by 8 to convert into Mbps is 1.265 Mbps**** >>> >>> **** >>> >>> Any ideas on how to explain the logic behind these results given the >>> testing setup described above?**** >>> >>> **** >>> >>> Cheers**** >>> >>> >>> >>> ------------------------------------------------------------------------------ >>> October Webinars: Code for Performance >>> Free Intel webinars can help you accelerate application performance. >>> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most >>> from >>> the latest Intel processors and coprocessors. See abstracts and register >>> > >>> >>> http://pubads.g.doubleclick.net/gampad/clk?id=60134071&iu=/4140/ostg.clktrk >>> _______________________________________________ >>> Iperf-users mailing list >>> Ipe...@li... >>> https://lists.sourceforge.net/lists/listinfo/iperf-users**** >>> >>> >>> >>> **** >>> >>> **** >>> >>> -- **** >>> >>> *Mohammad Hajjat***** >>> >>> *Ph.D. Student***** >>> >>> *Electrical and Computer Engineering***** >>> >>> *****Purdue**** University******* >>> >>> No virus found in this message. >>> Checked by AVG - www.avg.com >>> Version: 2013.0.3408 / Virus Database: 3222/6745 - Release Date: 10/12/13 >>> **** >>> >>> >>> >>> **** >>> >>> ** ** >>> >>> -- **** >>> >>> *Mohammad Hajjat***** >>> >>> *Ph.D. Student***** >>> >>> *Electrical and Computer Engineering***** >>> >>> *****Purdue**** University******* >>> >>> No virus found in this message. >>> Checked by AVG - www.avg.com >>> Version: 2013.0.3408 / Virus Database: 3222/6747 - Release Date: 10/13/13 >>> **** >>> >>> >> >> >> -- >> *Mohammad Hajjat* >> *Ph.D. Student* >> *Electrical and Computer Engineering* >> *Purdue University* >> >> >> ------------------------------------------------------------------------------ >> October Webinars: Code for Performance >> Free Intel webinars can help you accelerate application performance. >> Explore tips for MPI, OpenMP, advanced profiling, and more. Get the most >> from >> the latest Intel processors and coprocessors. See abstracts and register > >> >> http://pubads.g.doubleclick.net/gampad/clk?id=60134071&iu=/4140/ostg.clktrk >> _______________________________________________ >> Iperf-users mailing list >> Ipe...@li... >> https://lists.sourceforge.net/lists/listinfo/iperf-users >> >> > -- *Mohammad Hajjat* *Ph.D. Student* *Electrical and Computer Engineering* *Purdue University* |