Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!uunet!husc6!mit-eddie!uw-beaver!cornell!batcomputer!pyramid!voder!blia!forrest
From: forrest@blia.BLI.COM (Jon Forrest)
Newsgroups: comp.arch
Subject: Re: Sort Co-Processors (and database machines)
Message-ID: <3437@blia.BLI.COM>
Date: Mon, 12-Oct-87 13:36:53 EDT
Article-I.D.: blia.3437
Posted: Mon Oct 12 13:36:53 1987
Date-Received: Wed, 14-Oct-87 00:47:23 EDT
References: <112@sdeggo.UUCP> <1577@unc.cs.unc.edu>
Organization: Britton Lee, Berkeley, CA
Lines: 36
Keywords: sort co-processor, database machine

I thought I'd add a quick note about database machines since
special purpose hardware seems to be the topic of the week.

Database machines are appropriate when the amount of work neccessary
to perform the database task is considered "too high" to be performed
on a general purpose host or when data needs to be shared amoung
heterogenous environments. In the first case, we have seen that
our database machine doesn't do well when it is being used only as
a record server; the I/O time kills us. Most OS's can get the next
record from a sequential file faster that we can get the next tuple
(based on an ordering defined by the user) and send it over one
of the interfaces we support to a host, which then has to treat the
data the same as if it had come straight off a disk. However,
if the users wants data that is the result of a complex join,
perhaps involving a complex qualification, and wants the data sorted
then this is where our (or anyone else's) database machine is useful.
If the database machine could be more closely coupled to a host,
such as through a true co-processor, then the I/O bandwidth from the
database machine to the host would increase, resulting in higher
total "thoughput" being seen by users.

The other advantage to database machines is that data can be shared
by systems that either don't talk to each other cleanly or by systems
that don't have adequate data locking facilities. For example, a
VM/CMS user running an multi-user application might find that using
a database machine to store and maintain their data would be easier
than inventing ad-hoc methods for concurrency control. (I am told
that VM/CMS doesn't provide such a facility. I could be wrong).
In addition, it might be easier for a Unix (or VMS) Vax to share
data with a VM/CMS site by using a database machine than by linking
the two together.

Jon Forrest

ucbvax!mtxinu!blia!forrest
{pyramid|voder}!blia!forrest