From: bsa@kf8nh.wariat.org (Brandon S. Allbery) Subject: Re: SCSI Performance (Yet Again) Date: Tue, 24 Aug 1993 22:07:27 GMT
In article <110135@hydra.gatech.EDU> gt8134b@prism.gatech.EDU (Howlin' Bob) writes:
>bsa@kf8nh.wariat.org (Brandon S. Allbery) writes:
>>/dev/hda is still buffered... Linux doesn't have raw disk devices. I haven't
>>yet figured out why there's so much resistance to adding them.
>
>I don't see any resistance. Just because Linus hasn't done it doesn't
>mean he won't let them in. Everyone's busy. Would you write them?
>It shouldn't be too hard.
There are usually quite a few comments about how they aren't worth doing or
worth using when I bring them up... and I do know of a use. Database
managers. We've discussed ports of commercial software in this group before;
you aren't likely to see a serious commercial DBMS until raw devices are
supported. The reason being that buffer management which is optimized for one
method of access is often lousy for others (e.g. sequential vs. random) ---
the *ix buffer management method strikes a good balance between them for most
files, but leaves something to be desired for large databases, so most
commercial DBMSes have a means of running on a raw device *and implementing
their own buffer management*. The result is a considerable speed improvement,
unless raw devices are broken (I've seen that! --- a kernel patch to Plexus's
SVR2 increased database access speed about 5-fold). Or missing, in which case
you take a double speed hit because the OS and the DBMS are both doing
buffering... the OS doing so sub-optimally. (Yes, I know the commercial
DBMSes can run with the databases in ordinary disk files. But for *serious*
use they use raw devices.)
I keep threatening to implement raw devices --- eventually I will find the
time to do so.
++Brandon
-- Brandon S. Allbery kf8nh@kf8nh.ampr.org bsa@kf8nh.wariat.org "MSDOS didn't get as bad as it is overnight -- it took over ten years of careful development." ---dmeggins@aix1.uottawa.ca