Re: Multiprocessing: Oracle 7 & Solaris 2.3

From: James A. Lane <jl3a+_at_andrew.cmu.edu>
Date: Tue, 16 Aug 1994 12:25:03 -0400
Message-ID: <siICVTC00jWNEEC3Fv_at_andrew.cmu.edu>


>In V7.0 Oracle doesn't do any optimization *or* load balancing - the OS
>does! On Solaris and other System V4 variants - you can pbind individual
>processes onto specific processors, but Oracle still doesn't know, or care.

Not that you want to pbind processes except under very specific circustances, unless you want to bugger the scheduler and disrupt other activity.

>On the other hand V7.1 which is now hitting the streets (as it were), claims
>to perform query decomposition - splitting a query over the processors of
>a SMP box. The up/down side are yet to be timed, and it may take a rev or
>two to get the bugs out (they're already up to 7.1.3).

7.1.3 has limited parallelism which allows certain types of queries to be spread across CPUs. Also parallel loading and parallel index builds. Expect 7.1.5 to be the release to go to, if they get the query parallelism up to Informix standards. Hopefully parallel updates will also make a showing by then.

>The next step up is MPP on a Kendal , an NCUBE, or whatever!

Strongly disagree. MPP does not give you the benefits they claim. There are memory utilization problems and, worse, the I/O throughput on MPP systems is not anywhere near the level that you get on shrared memory multiprocessor. Your best bet at the high end is a good SMP implementation, which is currently Sequent or Pyramid. Also, MPP is on the rocks and looking for new markets, praying for openings in the database server market (where they get beat by good SMP systems), and high bandwidth video or simulation systems (because of the drying up of the defense and scientific markets). Received on Tue Aug 16 1994 - 18:25:03 CEST

Original text of this message