Re: Low AAC performance but only when tested through the file system
Jason Thorpe <thorpej <at> wasabisystems.com>
2003-12-02 06:13:11 GMT
On Dec 1, 2003, at 8:41 AM, Olaf Seibert wrote:
> I have installed a FreeBSD test installation on part of the disk, and
> seems to suffer the same fate as NetBSD. In systat iostat it also does
> not get more than about 60 write xfers/second, and an associated
> transfer speed of 4M/sec. So this tends to point to hardware again,
> since FreeBSD's AAC driver seems much further developed than NetBSD's.
> How that RawIO benchmark managed to get its much better results is
> a mystery to me.
What's probably happening is that the RawIO benchmark performs I/O in
such a way that it is able to make good use of the card's on-board
stripe cache. Presumably RawIO is doing large sequential writes, which
are broken up into 64k chunks, but possibly in a way that the card is
able to optimize (i.e. hold other stripe chunks in its stripe cache
until the entire stripe is present and eliminate the r/m/w cycle on the
On the other hand, the file system is issuing 64k I/Os, but not
necessarily all of them adjacent, so the card is probably having a
harder time optimizing the I/O with the stripe cache.
You could probably get some improvement by tweaking your stripe depth
on the card. You really want to be writing an entire stripe at a time,
so setting your stripe depth to (64k / ndisks-1) for RAID-5 should do
the trick (64k because that is the maximum I/O size the NetBSD kernel