Path: utzoo!utgpu!news-server.csri.toronto.edu!rpi!think.com!spool.mu.edu!uunet!xstor!iverson
From: iverson@xstor.com (Tim Iverson)
Newsgroups: comp.sys.ibm.pc.misc
Subject: Re: Disk seek optimisation
Message-ID: <1991Jun07.024728.29844@xstor.com>
Date: 7 Jun 91 02:47:28 GMT
Article-I.D.: xstor.1991Jun07.024728.29844
References: <1991Jun5.111750.951@csc.canterbury.ac.nz>
Organization: Storage Dimensions, Inc.
Lines: 22

In article <1991Jun5.111750.951@csc.canterbury.ac.nz> phys169@csc.canterbury.ac.nz writes:
>If software on a PC was to order read requests in such a way as to reduce head
>movement (I think some file servers use this, called "elevator optimisation" or
>similar), with the fast disks in use today, how much improvement is possible?

DOS is single tasking.  This means only one request for i/o happens at a
time.  If there is only one request, you have nothing optimize.

If you are talking about a multitasking system on a PC (not DOS), then yes,
you can improve by leaps and bounds over FIFO by using an algorythm that
reduces seek distances and/or latency time - latency is very hard
(impossible?) to reduce by scheduling and remain efficient.

Typically, head schedulers run O(n log n), but since you can save several
milliseconds by rescheduling just one movement correctly, it more than pays
for itself.  See your favorite OS text for an intro on head scheduling.

>Mark Aitchison, Physics, University of Canterbury, New Zealand.


- Tim Iverson
  iverson@xstor.com -/- uunet!xstor!iverson