[3872] in linux-scsi channel archive
Re: [BACK ON TOPIC] SCSI & IDE
daemon@ATHENA.MIT.EDU (Robert G. Brown)
Fri May 8 11:05:32 1998
Date: Fri, 8 May 1998 10:53:42 -0400 (EDT)
From: "Robert G. Brown" <rgb@phy.duke.edu>
To: Mr M S Aitchison <physmsa@cantua.canterbury.ac.nz>
cc: linux-smp@vger.rutgers.edu, linux-scsi@vger.rutgers.edu
In-Reply-To: <199805072251.KAA27297@cantua.canterbury.ac.nz>
On Fri, 8 May 1998, Mr M S Aitchison wrote:
> Suppose you start with a single *IDE disk, fast and cheap, on an SMP
> Linux system you assume won't need much file access 'cause it is going
> to be number crunching all day. Then you need more disk
> capacity/performance, so you add SCSI. Assuming you use hdparm for
> best IDE performance will it still drag down the performance of the
> system?
>
> I suppose it depends on what the EIDE is used for. If it is just
> "system files" (/var, /usr, etc) or just swap, could it impair CPU
> performance (given the Linux 2.0 system call "bottleneck"
> especially)??
>
> Would you REMOVE the IDE?
I have one system with both IDE and SCSI drives. I put the read-only,
write rarely, part of the system on the IDE drive (/, /usr, /etc and
even /tmp and /var). The SCSI drive (a Barracuda) is for serving and
work scratch space and swap. For the kind of code we are thus far
running on our Beowulf, the performance of this system is pretty much on
a par with all the other systems. If a parallel code were tightly
coupled, fine grained parallel, the configuration asymmetry might slow
down the whole calculation to the slowest common denominator, but I
think you are looking at <1% effects unless you are seriously disk I/O
bound. Even then, the running, nearly quiescent (except for the
calculation) system will only rarely read from the IDE disk (perhaps to
refresh a page) and should still get just about as good write
performance onto the SCSI channel as it would get if the IDE drive were
not there.
I wouldn't remove the IDE unless I had some direct evidence that it was
a problem. On the one I have, there is no such evidence.
rgb
Robert G. Brown http://www.phy.duke.edu/~rgb/
Duke University Dept. of Physics, Box 90305
Durham, N.C. 27708-0305
Phone: 1-919-660-2567 Fax: 919-660-2525 email:rgb@phy.duke.edu
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to majordomo@vger.rutgers.edu
