Date: Wed, 01 Jun 2005 11:23:06 -0600 From: Scott Long <scottl@samsco.org> To: Dan Nelson <dnelson@allantgroup.com> Cc: freebsd-stable@freebsd.org, Steven Hartland <killing@multiplay.co.uk> Subject: Re: good performing SCSI RAID5 ( was: asr ( 2015S ) support in 5.4 amd64? ) Message-ID: <429DEEFA.4040809@samsco.org> In-Reply-To: <20050601162914.GD90259@dan.emsphone.com> References: <006201c56633$5bc48c20$b3db87d4@multiplay.co.uk> <20050531233612.GD20906@tigerfish2.my.domain> <429D0C65.6010308@alumni.rice.edu> <003401c56696$0858c4d0$b3db87d4@multiplay.co.uk> <20050601150754.GE20906@tigerfish2.my.domain> <00b901c566bd$d48f2450$b3db87d4@multiplay.co.uk> <20050601162914.GD90259@dan.emsphone.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Dan Nelson wrote: > In the last episode (Jun 01), Steven Hartland said: > >>Thanks for that Bruce I'm quite surprised that these numbers are so >>low after playing with a cheapo hightech SATA controller which with >>the help of the guys on the list I was able to give out 200MB/s I >>really would expect the relatively expensive SCSI controllers to do >>significantly better especially as they have superior disks attached >>( 10K vs 7k2 ) and not performance which is well below ( 1/2 ) that >>expected of a single disk. > > > The faster rpms will get you more concurrent I/Os per second but won't > do as much for throughput. My asr 3200S cards got repurposed before I > could try them with 5.x, but with the 370F firmware I'm pretty sure I > was able to get more than 40MB/sec reads out of them on 4.x with 4-disk > RAID5 sets. Since the asr driver needs Giant, try a UP kernel and see > if it goes any faster. > A UP kernel won't help much because the interrupt handler will still fight with the CAMISR thread. Faster RPMs will get you faster throroughput and faster seeks. Don't forget that ATA/SATA plays fast-and-loose with the write cache and as a result is much less reliable in the event of a power failure or other problem. The cache tricks are what allow it to appear to have such good sequential write characteristics, though. Just like CPUs, benchmarking storage is all about customizing your hardware for the benchmark, not for the real world. Scott
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?429DEEFA.4040809>