Ok.... Whatever I've done SEEMS to be working, CPU is very low now on both Tx and Rx. That said, I have almost no idea what code changes are actually in place right now. I *think* it's a combination of Gerrits' nanosleep stuff in delay.ccp, Client.cpp, etc. - and the Conditional waits / usleep's in main.cpp, Reporter.cpp, etc.
I may wipe everything out and start over again with controlled tests between each change. It would be nice if *all* the necessary patches were integrated in one place - tarball, svn, etc. Any guidance on the proper / preferred method to address the high CPU load with UDP issue would be great. Thanks again! Gary -----Original Message----- From: Gary Gatten Sent: Friday, March 26, 2010 1:01 PM To: 'Tom Throckmorton' Cc: 'iperf-users@lists.sourceforge.net' Subject: RE: [Iperf-users] Gerrit Renker patch for high CPU The code in the svn/trunk dir SUX!!! CPU is maxed on both send and receive, at least on RHEL5. Also, when starting the client with "-r" and then aborting on the client (ctrl+c), the server side continues to transmit and the report echo'd to stdio is not formatted. I think this is new behavior. I haven't tested it on FBSD or Solaris yet, but I need it to work on RHEL5 so.... I'll keep reading, poking around, testing, etc... I really like this tool - hope I can get it to work. G -----Original Message----- From: Gary Gatten Sent: Friday, March 26, 2010 10:19 AM To: 'Tom Throckmorton' Cc: 'iperf-users@lists.sourceforge.net' Subject: RE: [Iperf-users] Gerrit Renker patch for high CPU Ah, I see! I had tweaked the code in Client.cpp myself, replacing "delay_loop( 0 );" with "usleep ( 1000 ); This definitely fixed my high cpu load - it barely uses any now for 100Mb/s streams. But, since I have almost no clue what I'm doing I'll integrate the SVN source you pointed out! Thanks so much for your help! Once I get this "working" I can deploy it to my HelpDesk staff and it will make my life a LITTLE tiny bit easier! G -----Original Message----- From: Tom Throckmorton [mailto:thr...@mcnc.org] Sent: Thursday, March 25, 2010 8:56 PM To: Gary Gatten Cc: 'iperf-users@lists.sourceforge.net' Subject: Re: [Iperf-users] Gerrit Renker patch for high CPU On 3/25/10 6:19 PM, Gary Gatten wrote: > Hello all, > > I've searched and read numerous posts on "high cpu for udp" issue. > Following a pretty long string I found and applied the "Ingo Molnar" > patch and it seemed to have work when iperf is RECEIVING traffic. > However, no love when it's sending traffic. > > So, I read some stuff that Gerrit Renker had a patch for the client > side (Client.cpp?) ; doing away with gettimeofday() and delay_loop(). > I can't find this patch anywhere! The link in this thread > http://www.mail-archive.com/iperf-users@lists.sourceforge.net/msg00028.html > doesn't work directly - and it appears to be the same location as the > "Ingo Molnar" patch - which isn't what I need. Hi Gary, I believe Gerrit merged that patch shortly after that thread - see: http://iperf.svn.sourceforge.net/viewvc/iperf/trunk/ChangeLog?view=log > 2008-05-09 Gerrit Renker <gren...@users.sourceforge.net> > > * replace costly gettimeofday() with nanonsleep() See: > https://sourceforge.net/tracker/index.php?func=detail&aid=1940009&group_id=128336&atid=711373 > So, where is this magic patch? I fetched the latest SVN and it's the > same as the 2.0.4 distro. You won't find it in the 2.0.4 tarball, only in trunk http://iperf.svn.sourceforge.net/viewvc/iperf?view=rev&revision=43 > I found one linux distro that has purportedly integrated these > patches into their package called 2.0.4-4, but I have not reviewed > that source yet... 2.0.4-4 in Fedora and EPEL do appear to contain the patches - see also: https://bugzilla.redhat.com/show_bug.cgi?id=506884 Cheers, -tt -- Tom Throckmorton MCNC - Advanced Services Development 3021 Cornwallis Road Research Triangle Park, NC 27709 919.248.1448 "Connecting North Carolina's future today" ------------------------------------------------------------------------------ Download Intel® Parallel Studio Eval Try the new software tools for yourself. Speed compiling, find bugs proactively, and fine-tune applications for parallel performance. See why Intel Parallel Studio got high marks during beta. http://p.sf.net/sfu/intel-sw-dev _______________________________________________ Iperf-users mailing list Iperf-users@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/iperf-users
