Date: 12 Nov 1998 00:31:42 -0600 From: Joel Ray Holveck <joelh@gnu.org> To: Greg Lehey <grog@lemis.com> Cc: john cooper <john@isi.co.jp>, hackers@FreeBSD.ORG, mike@smith.net.au, ticso@cicely.de Subject: Re: SCSI vs. DMA33.. Message-ID: <86k911ifqp.fsf@detlev.UUCP> In-Reply-To: Greg Lehey's message of "Wed, 11 Nov 1998 16:20:00 %2B1030" References: <98Nov11.134648jst.21907@ns.isi.co.jp> <19981111162000.O20374@freebie.lemis.com>
next in thread | previous in thread | raw e-mail | index | archive | help
>> The theoretical throughputs of 40MBs and 33MBs don't tell me a whole >> lot. I know SCSI was the choice for performance in the past, >> however I'm curious what others are seeing in actual usage these >> days. > I'm running 3.0-CURRENT (post-RELEASE), and as you can see Ultra DMA > is enabled on the IDE drives. Here's what I get transferring 32 MB > from each raw device: Were these filesystems in use at the time? About how many users/processes? dd's rarely will effectively emulate the access patterns of a real-world server. Numerous outstanding short requests scattered across the disk is more common. Several simultanious worldstones may model real-world performance better. I know that you know this, and most of the -hackers list as well, but I'd like to stress that these methodologies are all quick n' dirty. Seek times, onboard transfer buffers, and other aspects of the drives will throw these numbers off. A good methodology would be using identical drive controllers and mechanisms on the IDE and SCSI. (Naturally, ad-hoc is all we can provide right now.) > Looks good for the IDE drives, doesn't it? They say, though, that > SCSI drives work better with multiple requests outstanding... It is my understanding that the multiple request issue is primarily concerned with multiple requests on the same drive. You may want to try this again with some more dd's with skip= arguments to see how much tagged command queueing and device detachment help. Does anybody know how many drives support the SCSI device-to-device copy capability? This would probably only be useful in dd's, but I've considered implementing it. Happy hacking, joelh -- Joel Ray Holveck - joelh@gnu.org Fourth law of programming: Anything that can go wrong wi sendmail: segmentation violation - core dumped To Unsubscribe: send mail to majordomo@FreeBSD.org with "unsubscribe freebsd-hackers" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?86k911ifqp.fsf>