Hi, [email protected] (2012-08-28 at 2025.46 +0200): > Ah, yeah, thanks. > > Please excuse me if I now dump this here, but I have a hard time > figuring this one out. I have a 26MiB strace log just from ~15 seconds > running and tried to analyze it with all my might. Though, I feel like > poking into the dark... > > Be warned, this is 26MiB+: > strace -txo strace_output sawfish-config > http://www.bonsaimind.org/tmp/strace_output
Next time gzip it, it goes down to less than 800KB. ;] > I have also attached the output of 'strace -c sawfish-config'. I can see > nothing too suspicious except these 'gettimeofday' and 'clock_gettime' > calls...which are quite often in my opinion. Writes do not seem to be > that frequent (roughly 320 calls), some of them seem to happen with > quite big packages, but it still doesn't look like that this would > explain the big disk activity I'm seeing. There are thousands of 'read'. The one that appears a lot uses fd 10, 'grep "read(10" strace_output | wc -l' gives 116465, the file has 261125 lines. Looking up what has generated a 10 gives some 'open's (that are closed quickly) and 'inotify_init1'. Those read calls have '\x73\x74\x72\x61\x63\x65\x5f\x6f\x75\x74\x70\x75\x74' which is the string 'strace_output'. My take is something is looking for changed files (check the 'inotify_add_watch' calls), and you are changing a file just where that thing is looking at. Could you try making redirections to /tmp/strace_output and check again? Maybe something (gtk inner guts?) is causing part of the IO noise. GSR -- Sawfish ML
