Date: Tue, 10 Nov 1998 22:58:03 -0800 (PST) From: Matthew Dillon <dillon@apollo.backplane.com> To: john cooper <john@isi.co.jp> Cc: grog@lemis.com, hackers@FreeBSD.ORG, mike@smith.net.au, ticso@cicely.de, john@isi.co.jp Subject: Re: SCSI vs. DMA33.. Message-ID: <199811110658.WAA18122@apollo.backplane.com> References: <98Nov11.134648jst.21907@ns.isi.co.jp>
next in thread | previous in thread | raw e-mail | index | archive | help
The biggest issue with IDE is that it's a serialized interface. You
can only run one command at a time to one device at a time and you
have to wait for it to complete before you can run another command
to the same device or another device. The throughput is therefore
irrelevant, since most disks can't do more then 9 MBytes/sec off
the platter anyway (note: the latest bleeding edge IBM drives can
apparently do 20 MBytes/sec or better off the platter). IDE also isn't
too hot when it comes to bad sector remapping or multiple devices,
which is why most PC's come with two separate IDE busses these days. One
for the CDRom, and one for the disk, nor is IDE necessarily reliable
when sharing a bus in a master-slave configuration, even if both drives
are the same brand.
On the otherhand, a 40 MByte/sec SCSI bus with four 9 MByte/sec drives
on it can actually be doing 36 MBytes/sec on the bus.
Since, typically, anything approaching that kind of bandwidth is also
an indication for the need for performance, it seems silly to me
to even begin to compare IDE with SCSI no matter what the DMA transfer
bandwidth of the IDE device is.
I suppose if you never needed to seek the drive it might matter, but the
moment you start needing to seek the drive the platter bandwidth goes
to pot. The best a fully saturated randomly seeking disk can do is
typically less then 2 MBytes/sec using 16K reads, and it doesn't get
much better with larger reads. It takes truely large reads (256 KBytes
or larger) between seeks to even approach the platter's bandwidth. The
size of the disk in the seeking case is irrelevant, really, because
voice-coil technology has not gone through the same insane technology
leaps that the disk heads have gone through.
IDE is still useful when you don't need performance, and I would say
that 80% of the server installations these days fall into the 'don't
need the performance' category. My home machines are a mix of IDE and
SCSI, but all of the rack mount FreeBSD boxes in BEST's machine room
are SCSI-only. If you do need performance, you go with SCSI.
-Matt
:
:Hi,
: Just wondering if anyone has any _objective_ opinion on
:the performance of say wide SCSI2 vs. DMA33 IDE drives [running
:on contemporary motherboards]. 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.
:
:Thanks,
:
:-john
:
:To Unsubscribe: send mail to majordomo@FreeBSD.org
:with "unsubscribe freebsd-hackers" in the body of the message
Matthew Dillon Engineering, HiWay Technologies, Inc. & BEST Internet
Communications & God knows what else.
<dillon@backplane.com> (Please include original email in any response)
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?199811110658.WAA18122>