SUMMARY: amount of data read per read request on DU4.0B

From: VAN DE PERRE Jochen <Jochen.Van.De.Perre_at_is.belgacom.be>
Date: Thu, 30 Jul 1998 09:20:34 +0200

Thanks to:
        George Guethlein (GGuethlein_at_GiantOfMaryland.com)
        Stephen Mullin (Stephen.Mullin_at_gecits.ge.com)


This parameter (KB/read request) can be set in two ways (may
come inhandy when choosing how to setup an oracle DB and
finetuning it)...

1) If you use LSM and stripesets, you can specify a stripewidth
(default=64k) when creating these stripesets. A "volprint -htA | more"
command gives you a full review of your LSM configuration. Look for
"STRIPE". These are your stripesets, and the stripewidth should be
mentioned as well.

I did not find any "STRIPE" (we do not use LSM stripesets), so my
only option (!) was to use possibility 2 (below).

2) If you have the Storage Works technology, you can set a parameter
called "MAXIMUM_CACHED_TRANSFER_SIZE" between 1 and 1024
blocks, when you create a unit (stripeset, mirrorset or raidset) with
the storage works command language. To verify your current settings,
use "show raidsets (or stripesets, or mirrorsets) full".

 
Stephen Mullin told me that a good tuning effort for SAP oracle
is to build raid3/5 units with a chunksize of 64kb, an then add the
following to your sysconfigtab:

advfs:
        AdvfsCacheMaxPercent = 1
        AdvfsMaxDevQueueLength = 32
        AdvfsFavorBlockingQueue = 0

but, as said, this is for SAP DB's.

\\
Jochen Van de Perre
SAP & UNIX system administrator
Belgacom Directory Services
Tel: 02/702.70.45
//

Weekly quote:
"640K ought to be enough for anybody" - Bill Gates, 1981.
Received on Thu Jul 30 1998 - 07:39:03 NZST

This archive was generated by hypermail 2.4.0 : Wed Nov 08 2023 - 11:53:38 NZDT