On my SMP system, sometimes the whole system (games, KDE, sound) freezes for an interval of about 1-3 seconds, then runs on. At the same time, the proc graph (showing number of forks/new processes as far as I understand it) in gkrellm makes a leap before it gradually goes down again. Bringing one of the cores under constant load (via "cat /dev/urand > /dev/null") seems not to trigger this, but e.g. playing neverwinter nights seems to increase the frequency of the freezes. However, the lags are also present when the computer is idle. The computer is completely new, so I can't pinpoint some hardware that could be made responsible. specs: 2.6.26-gentoo-r3 #11 SMP Fri Dec 5 13:52:03 CET 2008 x86_64 AMD Phenom(tm) 9350e Quad-Core Processor AuthenticAMD GNU/Linux ATI 790GX chipset, onboard graphics used Samsung F1 SATA drive ati-drivers-8.542 X Window System Version 1.3.0 I know this isn't too much info, but maybe you can help me track down the issue. I'll attach the output of emerge.info. If you need additional info, I'm happy to help Reproducible: Couldn't Reproduce
Created attachment 174580 [details] emerge --info output
If those lags occur for a few seconds and coincide with a jump in the number of processes, we should have a good clue by learning what those new processes are... Try running a script that runs 'ps axf' every second and appends the output to a file. Kill the script shortly after a lag occurs, and hopefully we'll see the burst of new processes towards the end of the file.
Created attachment 174663 [details] the last ~30 seconds of "ps axf" output The "TICK" strings were inserted by me to distinguish the individual outputs. I ran Enemy Territory to help recognize the freezes (and it also looks as if games increase the probability of encountering a freeze). If you need more output, I just cut out the last seconds as the whole file is more than 9 MB large. Maybe you can make something out of that info. Thanks for the effort :-)
Created attachment 174748 [details] Another process dump while running nwn Here is the process list of another lag occurence, this time while playing Neverwinter Nights (it doesn't just happen when playing games, but they make the lags most obvious). The lag occured several seconds before the last entry, because I first have to stop the game in order to stop the script. Enjoy! :-)
Sorry for the spam, but I've got some disappointing results: I did some rudimentary analysis of the process lists. first I erased all process numbers that were already there at the beginning of the list, but ended up with (almost) no new process numbers, only ps seems to get new ones from time to time. To verify that I grepped the lines where my TICK entry was written to, and found out that the line numbers are strictly linear, so there were no new processes started, at least none ps could log. So this seems to be a rather dead end. Other ideas? Something I have overlooked, some information I haven't used?
This may have to do with your ati-drivers. As you 're using stable amd64, why don't you try the stable ati-drivers(ati-drivers/ati-drivers-8.471.3.ebuild)? Maybe you can also try the latest ati-drivers but, you Xorg system is the stable old 1.3.0. I don't believe that this is gentoo-sources specific.
I don't use the stable driver because they doesn't compile with the stable gentoo-sources, whereas the testing driver does. I know my xorg is somewhat antique. But as the newer versions are still marked as testing, I'd have to keyword all relevant packages, and normally I like to keyword packages with an exact version so I don't one testing version after the other. For Xorg I did it once, and nearly died while iteratively trying to compile, getting next required package with version, put it into package.keywords, trying to compile again, etc. But if there is a chance that it helps I can try.
(In reply to comment #7) > I don't use the stable driver because they doesn't compile with the stable > gentoo-sources, whereas the testing driver does. > > I know my xorg is somewhat antique. But as the newer versions are still marked > as testing, I'd have to keyword all relevant packages, and normally I like to > keyword packages with an exact version so I don't one testing version after the > other. For Xorg I did it once, and nearly died while iteratively trying to > compile, getting next required package with version, put it into > package.keywords, trying to compile again, etc. Can you at least try the latest ati-drivers(ati-drivers/ati-drivers-8.552-r2) ? > > But if there is a chance that it helps I can try. >
OK, upgrading to Xorg 1.5.2 and ati-drivers 8.542 helped, the lags are gone. I now have some minor issues, but I'll tackle them when I have some more idle time. Thanks to all :-)