Re: [PATCH] blk queue io tracing support

[Jens Axboe <axboe@suse.de>]

On Wed, Aug 24 2005, Nathan Scott wrote:
> On Tue, Aug 23, 2005 at 02:32:36PM +0200, Jens Axboe wrote:
> > Hi,
> > 
> > This is a little something I have played with. It allows you to see
> > exactly what is going on in the block layer for a given queue. Currently
> > it can logs request queueing and building, dispatches, requeues, and
> > completions.
> 
> Ah, fabulous.  Thanks Jens!
> 
> > +	t.magic		= BLK_IO_TRACE_MAGIC | BLK_IO_TRACE_VERSION;
> > +	t.sequence	= atomic_add_return(1, &bt->sequence);
> > +	t.time		= sched_clock() / 1000;
> 
> Wouldn't it be better to pass out the highest precision available here
> & then do the conversion in userspace instead?  I guess one might want
> that little bit more for a RAM disk or something ... actually, talking
> to one of the SGI people here with alot of experience on IRIX with a
> similar facility, the msec resolution there is apparently sometimes an
> issue already with fast storage.

This isn't msec precision, it's usec. sched_clock() is in ns! I already
decided that msec is too coarse, but usec _should_ be enough.

-- 
Jens Axboe