Date: Sat, 23 Nov 2002 18:57:40 +0000 From: Pete French <pfrench@firstcallgroup.co.uk> To: stable@FreeBSD.ORG Subject: RAID performance problems Message-ID: <E18FfTI-0004KM-00@mailhost.firstcallgroup.co.uk>
next in thread | raw e-mail | index | archive | help
I wonder if anyone can shed any light on the following, as I am starting to come to the conclusion that its possibly some BSD driver issue. I have been testing the performance of the various drives on my system using bonnie, just out of curiosity. The system has 3 different drives on it: 1) a 15k RPM drive attached directly to an Adaptec SCSI controller 2) a RAID-1 mirrored pair of 10k RPS drives 3) a RAID-5 set of 5 7200 RPM drives All SCSI buses are 40MB/sec. Both the RAID drives are being run from the same Compaq 4200 RAID controller. Both the RAID controller and the Adaptec SCSI card are on the same PCI bus. So, what I expected was that the driect drive would be fastest, followed by the RAID-1, then the RAID-5 (at least as far as write performance goes) The actual results are that both the RAID drives have identical performance! The direct drive is 4 times faster than these. This was a bit of a shock to me. Especially ater all the effort of utting in thr 10k RPM drives to try and speed that disc up. The obvious conclusion is that somewhere there is some component common to both RAID drives which is limiting the performance. The bottleneck cannot be the PCI bus to the card as the direct drive is shifting 4 times the data on the same bus. It also cannot be the SCSI bus to the drives by my reasoning - even assuming that the controller aggregates the writes so that RAID-5 and RAID-1 are taking the same bandwidth on the bus (which is necessary if they are to perform the same with the bus being the bottleneck right?) then the bus is operating 4x slower than the direct bus. The Compaq card supports 40MB/s bus speeds according to the documentation - plus the bus is terminated correctly as plugging the array into the SCSI card shows all the drives up as 40MB/s So I'm left with one of two conclusions a) This is a limitation in the processing power of the RAID card. This seems very unlikely - the work needed to be done by the controller for RAID-1 is far less than that which needs to be done for RAID-5 (it would take a complete muppet to design a controller where this wasnt the case wouldnt it ?) and this is one of Compaq's meatier controllers afer all. b) This is some strange limitation in the speed at which FreeBSD's device driver for Compaq controllers can talk to the actual device. Its the fact that I have arrived at conclusion b) as a possibility that makes me ask on here. Does anyone have any comments or thoughts on this ? I know there are people around who can argue at great length about RAID performance after all :-) -pcf. [puzzled] To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-stable" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?E18FfTI-0004KM-00>