On Saturday 18 February 2006 18:54, Lee Revell wrote: > On Sat, 2006-02-18 at 00:47 -0800, Kevin Cosgrove wrote: > > On 18 February 2006 at 0:15, Lee Revell <rlrevell@xxxxxxxxxxx> wrote: > > > This helps a lot: > > > echo 64 > /sys/block/hdc/queue/max_sectors_kb > > > (replace hdc with your drive, repeat for all of them) > > > > Interesting. What does that do? > > It limits the amount of disk IO that can be "in flight" at once, which > limits the amount of work the IDE driver will do in a single interrupt, > which improves the worst case disk IO induced latency. > > It's not going to help with an -rt kernel, only mainline. > > I'd be interested to see what results people get... > > Lee I suppose it also work with SATA drives, doesn't it? At least /sys/block/sda/queue/max_sectors_kb is available here. I just tried with Ubuntu's dapper's stock 2.6.15 kernel. Jackd runs at 44100Hz samplerate and buffersize 256 samples (5.8ms periods). Hardware is an AOpen i915GMm-HFS board with a Pentium M 730 (1.6 GHz) and a RME Digi96/8 PAD. Default max_sectors_kb is 1024, giving me casual xruns of up to 16 ms on an idle machine. With max_sectors_kb at 64, xruns reduce to max. 5 ms. Setting max_sectors_kb smaller than 64 seems to give no further improvement. So it definitely helps, but not enough to go without -rt for me. Wolfgang
