From: Stefan R. <st...@s5...> - 2011-07-12 06:58:02
|
On Jul 12 Arnold Krille wrote: > On Tuesday 12 July 2011 00:09:34 Stefan Richter wrote: > > On Jul 11 Andy Campbell wrote: > > > I believe that every time I refresh latencytop after an > > > unhandled XRUN, I see a "process" named jbd2/sda4-8 with >40 ms > > > latency. [...] > > jbd2 is a kernel thread; it deals with the journal on sda4. jbd2 is used > > by the ext4 and ocfs2 filesystems. > > > > Its development mailinglist is lin...@vg.... It's probably > > best to let the developers know that the journalling process appears to > > heavily impact audio latency. > > Better not write the kernel devs about that without good prove. ext[2-4] are > working very well for audio for a lot of people. It would be surprising if > this was really the cause for problems no one with similar setups and usage > encountered before... I suppose if Andy switches to CONFIG_PREEMPT=y, any latency in the jbd2 thread should become a non-issue. AFAIU jackd/ffado userspace threads are going to be able to preempt the jbd2 kernel thread with CONFIG_PREEMPT=y. If a few common distributions install "/boot/config-$(uname -r)" --- or "/lib/modules/$(uname -r)/{build,source}/.config" ? --- it would be nice to have ffado-diag check for CONFIG_PREEMPT=y or CONFIG_PREEMPT_RT=y there. Perhaps I'll send a patch. -- Stefan Richter -=====-==-== -=== -==-- http://arcgraph.de/sr/ |