Date: Thu, 1 May 1997 18:15:22 -0700 From: "Jin Guojun[ITG]" <jin@george.lbl.gov> To: craig@tuna.progroup.com Cc: hardware@FreeBSD.ORG Subject: Re: 53C875 UW SCSI-3 adapter Message-ID: <199705020115.SAA17849@george.lbl.gov>
next in thread | raw e-mail | index | archive | help
}> Under both 2.1.7 and 2.2.1, the 53C875 ultra-wide SCSI-III host adapter, made }> by J Bond Computer Systems (JDC5075), has the same performance as the 53C825, }> which is just a wide SCSI host adapter, about 18 MBytes/sec. }> The disk drive I used are Seagate BARRACUDA 9 (ST19171W) , which seagate }> claims that can go 40 MBytes in sychronous-transfer mode. } }How many drives do you have on the scsi buss? } }What did you use to test the speed? } }18MB/sec sounds pretty good to me. 2-4-8 disk drives are daisy chained. Basically, 2 of ST32171W drives are enough to saturate the SCSI bus (10.5 MBps per disk drive). using 4 or 8 drives are just for further testing. Also, Seagate Hawk disks (6.28 MBps per disk drive) have been used for the same testing. Aggregate thoroughput are similar -- 17.8 MBytes/sec. Therefore, the problem is either the device driver or one of the drive/adapter not fit the spec. (See below). }> }> Has someone experimented the throughput of 53C875 ultra-wide SCSI-III host }> adapter with any wide or ultra-wide drives? }> }> -Jin }> }> } }I am using a 2940UW with 4 ST32171W drives on a Solaris 2.5.1 x86, with }1 drive as the boot drive, and the other 3 striped (32k) using disksuite }4.0. Bonnie gives me about 12MB/sec max on the striped drives. I am sure this Solaris 2.5.1 x86 SCSI driver's problem. The same cable, drives, and SCSI host adapter for above testing are used for testing on the same machine with Solaris 2.5.1 x86. The speed is: 8.4 MBps for single ST19171W dirve 9.3 MBps for more than one ST19171W dirves I would believe that SCSI driver under Solaris 2.5.1 x86 is not compliant with SCSI-III specification. It can only handle 10 MBps spec. Otherwise, how to explain the different behavior between BSD and Solaris kernel with exact same hardware components. BTW, all my test is done by raw device I/O. When you mentioned to use disksuite 4.0, I will guess the 12MB/sec max is via the file system stripping, or not? -Jin
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199705020115.SAA17849>