seeing if i can get in on this. probably not, but i wanted to try. sorry for the nuisance, all.
> On Sep 9, 2008, at 4:55 PM, Eugene Gavrilov <[email protected]> wrote: > Hi Jeff, > >> On 09.09.2008, at 19:54, Jeff Moore wrote: >> >> The telemetry here looks problematic. If this is really from 5-10 minutes >> into the IO, there is no way that the rate scalar should still be 1.0. There >> just isn't any real hardware that runs with a perfect rate scalar. >> Typically, this means that the time stamps coming from the driver are >> causing the HAL to reset it's clock all the time. >> Further, there appears to be a wobble in IO cycle pattern. Normally you'd >> expect the difference between successive cycles (the Offset column) to be >> pretty much the same each time. In addition, there appears to be a huge >> scheduling latency (1.256ms) for a few of the cycles. So, in addition to bad >> time stamps, this probably means your driver is doing way too much work on >> the primary interrupt and that extra work is impacting the rest of the >> system adversely in the form of additional scheduling latency. >> So, there is definitely evidence of a badly behaving driver here. I'd look >> first at why your driver seems to be causing so much scheduling latency. >> That's a fatal issue since it impacts everything in the system. > > Thank you very much for your hints! I figured out that the huge scheduling > latency was caused by a 3-rd party network driver, I disabled the network and > it disappeared. > > However, 'rate' was still 1.00000, and I re-check my interrupt handler: there > was a major issue with it, and takeTimeStamp() was not called properly. After > I fixed this, rate started to deviate from 1.0 (0.9997, 1.00012), the problem > with clicks and sync disappeared, and I also noticed that this time 'Zero: > samp / host' columns ware no longer empty. > > I once again appreciate your help and thank you for fast response! > > Regards, > Eugene Gavrilov > Senior Software Engineer, > CEntrance, Inc. > _______________________________________________ > Do not post admin requests to the list. They will be ignored. > Coreaudio-api mailing list ([email protected]) > Help/Unsubscribe/Update your Subscription: > https://lists.apple.com/mailman/options/coreaudio-api/egavrilov%40centrance.com > > This email sent to [email protected] _______________________________________________ Do not post admin requests to the list. They will be ignored. Coreaudio-api mailing list ([email protected]) Help/Unsubscribe/Update your Subscription: https://lists.apple.com/mailman/options/coreaudio-api/archive%40mail-archive.com This email sent to [email protected]
