sklower@ucbvax.UUCP (09/11/83)
>From GEOFF5@SRI-CSL Sun Sep 11 00:39:47 1983
I just got in the mail my copy of the proceedings of the Spring 1983
DECUS meeting. On page 339 is the paper I mentioned in a previous posting
to this bulliten board on measurements of disk performance. It is by
Richard Wrenn of Washington University. I just reread it and it
is a very thoughtful analysis of the real issues in disk speed and
system performance. I highly reccomend it to anyone about to purchase
new hardware (after you've listened to all the hype from the salesmen.)
It reports on both simulations and measurements of performance of
RM80, R80 (on 11/730), SI9751 (Eagle), and UDA50-RA81.
Issues of transfer size, disk fragmentation, and requests from
high priority tasks (eg real-time) are all adressed. The issue
of whether you are really disk bound (rather than CPU bound) is
also adressed.
The one gap in the analysis was that the author had no means to test
the HSC50-RA81 combination. Perhaps by the Fall DECUS enough of these
will be in the field to yield some results on performance. Anyone got
any hands-on experience with the HSC50 yet?
Also unknown (to me at least) is how the performance figures would
look under Unix, since the IO system and disk structure there would
presumably give different results. Anyone done any testing of these
things with Unix?
Rg
-------