[5099] in linux-scsi channel archive
Re: Linux as a SCSI _Target_ device?
daemon@ATHENA.MIT.EDU (MOLNAR Ingo)
Thu Nov 12 00:19:31 1998
Date: Wed, 11 Nov 1998 23:48:22 +0100 (CET)
From: MOLNAR Ingo <mingo@chiara.csoma.elte.hu>
To: Jeff Noxon <jeff@planetfall.com>
cc: linux-kernel@vger.rutgers.edu, linux-scsi@vger.rutgers.edu,
linux-raid@vger.rutgers.edu
In-Reply-To: <19981111144602.B20117@planetfall.com>
On Wed, 11 Nov 1998, Jeff Noxon wrote:
> One big problem is stuffing a lot of IDE interfaces in a linux box.
> The hardware would be trivial (i.e. an 8-channel PCI IDE controller).
> But it's not exactly an off-the-shelf thing.
If you are building an 'off-the-shelf' Linux based hardware-RAID box,
you'll be better off with SCSI. No need for Ultra, just Wide and lots of
cards, lots of disks. Casing is not cheap usually, especially if it's
hot-swap, but it's such a simple component that i suspect it's just
because it's a server niche ... Anyway, Linux-RAID scales up to a few 10s
of MB/sec IO bandwith. To go near 100MB/sec it has to be a careful pick as
it's near what todays PII based system can provide ... in theory. A single
(dedicated) A dual-PII can deal with this, the MMX-XOR routine saturates
memory bandwith up to ~300MB/sec realistically. (of course at the expense
of CPU time, but this isnt a problem for a dedicated box) I'd worry more
about PCI bandwith, and about those custom 'generic SCSI device' cards.
(with such a solution, IO bandwith goes through the PCI bus twice, so we
have a max about ~100MB/sec)
-- mingo
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.rutgers.edu
