Date: 07 Oct 1998 21:31:50 +0300 From: Osma Ahvenlampi <oa@spray.fi> To: Michael Meissner <meissner@cygnus.com> Cc: Paul Haigh <paul@nailed.demon.co.uk>, "aic7xxx@freebsd.org" <aic7xxx@FreeBSD.ORG> Subject: Re: Synchronous/Asynchronous and Disconnect/Connect? Message-ID: <m3r9wkgr55.fsf@ws142.spray.fi> In-Reply-To: Michael Meissner's message of "Wed, 7 Oct 1998 08:16:01 -0400" References: <01BDF175.B29F64A0.paul@nailed.demon.co.uk> <19981007081601.A25700@tiktok.cygnus.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Michael Meissner <meissner@cygnus.com> writes:
> Basically Async is the original SCSI-1, Sync is newer and faster ('Fast' is the
> 20 Mb/s improvement, 'Ultra' is the 40 Mb/s bus speed). Disconnect is for slow
> devices not to hog the scsi bus.
Where an extreme example of a "slow device" is something like a tape
drive rewinding. Without disconnect, nothing else could be done on the
bus until the tape had rewound itself and the command had
returned. With disconnect, the initiator can tell the target to let go
of the bus and send a reply back once they're finished, so the
initiator can use the bus with other devices meanwhile.
Also, Fast/Ultra are bus clock speeds, whereas the transfer speed is a
function of bus clock speed and bus width:
bus speed narrow wide
SCSI-1 5 MHz 5 MB/s N/A
Fast 10 MHz 10 MB/s 20 MB/s
Ultra 20 MHz 20 MB/s 40 MB/s
Ultra2 40 MHz 40 MB/s 80 MB/s
Asynchronous transfer are without a clock signal, and are always
somewhere below the maximum synchronous speeds. For SCSI-1, I never
saw async speeds better than 3.7MB/s or so even on the best
implementations (which were very close to 5MB/s in synchronous mode)
Incidentally, 10 Mb/s == 1.25 MB/s, since b == bit. SCSI has parallel data
lines (for now, anyway), so its speed is never measured in bits per
second, unlike the serial connections like Ethernet, Fibre or USB.
> No, what you are missing is the scsi bus speed (40 Mb/s) is the maximum
> possible data movement for all devices connected to the bus (ie, you might be
> able to reach 40 Mb/s if you have 3-4 cheetah's all tallking at the same time).
> It's the guaranteed not to exceed speed, not the average speed you will see.
Well, in benchmarks ran on Beowulf data servers (I've seen the results
somewhere on beowulf.org, although now I can find data only for IDE
disks), it became pretty apparent that more than 3 disks per
controller, and the sustained speed flattened significantly before the
theoretical maximum bus capacity. 4 Cheetahs on the same bus probably
won't get you 40 MB/s, while the same drives on two Ultra-Wide
controllers probably would surpass 40MB/s aggregate speed in a fast
machine.
--
Paul's Law: You can't fall off the floor.
Osma Ahvenlampi <oa@spray.fi>
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-aic7xxx" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?m3r9wkgr55.fsf>