From: news@comlab.ox.ac.uk Subject: Re: hard disk performance Date: 21 May 1992 14:37:39 GMT
In article <May.20.17.28.38.1992.11957@dartagnan.rutgers.edu> hedrick@dartagnan.rutgers.edu (Charles Hedrick) writes:
>>MFM disk: DOS = 229 kb/sec Linux = 94.4 kb/sec
>>SCSI disk: DOS = 2847 kb/sec Linux = 90.2 kb/sec
>
>We've had postings like this before. As far as I can tell, the
>current hd code can't keep up with a reasonably fast controller. Thus
>I suspect you're getting one or two blocks per rotation on your SCSI
>disk. Using interleaving would make a big improvement. I still think
>this is the way to go until people have had a chance to optimize
>things a bit more. There seems to be a horror or interleaves. Yes,
>it's good to run with 1:1 interleave. But if your driver can't keep
>up with 1:1 interleave, it's a lot better to use 1:4 interleave than
>to get one block per rotation.
I've considered this step. Has anyone managed to hack mkfs / the
kernel to allow interleave on IDE drives? (I am assuming from
hearsay that since you can't low-level format IDE drives
you can't interleave onto the disc that way.)
Andrew
Andrew Stevens
Programmming Research Group JANET: Andrew.Stevens@uk.ac.oxford.prg
11 Keble Road, Oxford, England UUCP: ...!uunet!mcvax!ukc!ox-prg!as
OX1 3QD +44 0865 272563