I've just uploaded a new ag-rat package to the UQVislab maverick repo. Its based on a slightly earlier version (as was used in Ubuntu 10.4) and does not exhibit the memory leak problem.
chris On 28/11/2010, at 4:17 PM, Christoph Willing wrote: > I can confirm this problem. It appears to involve a small memory > leak in rat itself when using an ALSA device whenever Talk is > enabled. This becomes a major memory leak when Pulse audio is > involved - a 0.1% increase every few seconds in the rat-*-media > process, as well as a jump in cpu usage from 1 or 2% to 36%. > > It was previously e.g. Ubuntu 10.4, possible to use OSS instead of > ALSA (and therefore have no memory problem) but OSS is no longer > available in Ubuntu 10.10. > > Here's a summary of test results using code form the svn trunk (r4891) > > On a system with no Pulse Audio, both OSS and ALSA available: > OSS selected - uses total 1% cpu, 0.3% mem., irrespective of Talk > being on or off > ALSA selected - uses total 1% cpu, 0.3% mem., adds about 0.1% every > 30mins > > On a system with Pulse Audio & ALSA, but no OSS (Ubuntu 10.10) > Talk off - uses total 2% cpu, 0.2% mem usage at start, 0.8% after > ~2hours > Talk on - uses total 36% cpu, 0.2% mem usage at start, increasing > by 0.1% every 3 seconds > i.e. 30% mem usage after 10 minutes > on the test machine. > > I see from the rat svn repo that there's been no change in the > source code for a long time, so the problem must have been around > for a long time; maybe it hasn't been noticed because most users > have previously chosen to run rat with OSS rather than ALSA+Pulse, > or perhaps the current version of Pulse particularly exercises the > faulty code. Also, on the test machine at least, it requires using > rat with Talk on for about 30 minutes before the problem would be > noticed (unless you just happen to be carefully watching top at the > same time). > > > chris > > > On 25/11/2010, at 8:11 PM, Philippe d'Anfray wrote: > >> Bonjour, >> >> We used our ubuntu 10.10 server to retransmit a seminar last week. >> >> But RAT was "eating" all the available memory and eventually crashed >> every 10-15 minutes. >> >> syslog:Nov 18 17:02:32 aristotest1 kernel: [31358.739815] Out of >> memory: kill process 3720 (rat-4.4.01) score 1977288 or a child >> syslog:Nov 18 17:02:32 aristotest1 kernel: [31358.739820] Killed >> process 3721 (rat-4.4.01-ui) vsz:44748kB, anon-rss:2044kB, file-rss: >> 12kB >> syslog:Nov 18 17:02:32 aristotest1 kernel: [31358.800101] >> rat-4.4.01-ui: page allocation failure. order:0, mode:0x201da >> syslog:Nov 18 17:02:32 aristotest1 kernel: [31358.800106] Pid: >> 3721, comm: rat-4.4.01-ui Not tainted 2.6.35-22-generic #34-Ubuntu >> >> Has anyone else experienced a similar problem? >> >> Cordialement >> >> Philippe >> >> >> >> >> <rat.jpg><Philippe_d-Anfray.vcf> > > Christoph Willing +61 7 3365 8316 > QCIF Access Grid Manager > University of Queensland > Christoph Willing +61 7 3365 8316 QCIF Access Grid Manager University of Queensland
