HP3000-L Archives

January 2001, Week 1

HP3000-L@RAVEN.UTC.EDU

Options: Use Monospaced Font
Show Text Part by Default
Show All Mail Headers

Message: [<< First] [< Prev] [Next >] [Last >>]
Topic: [<< First] [< Prev] [Next >] [Last >>]
Author: [<< First] [< Prev] [Next >] [Last >>]

Print Reply
Subject:
From:
THOMAS COOK <[log in to unmask]>
Reply To:
THOMAS COOK <[log in to unmask]>
Date:
Wed, 3 Jan 2001 20:30:46 -0600
Content-Type:
text/plain
Parts/Attachments:
text/plain (64 lines)
I second Neil's suggestion.

Question, why is it that you wish to do this?

I am no certain of any real gain, what are you looking for? I mean
'prefetch' occurs whether the flag is on or not. The OS it self does
prefetch to a significant degree.

The only situation that I have been able to prove that prefetch does not
either work or is very inefficient is in backward IMAGE reads.

If there is no overwhelming need, you should never do a backward serial or
chained read. Unless you are only wanting the last 'x' occurrences or such
then these type reads should not be encouraged.

The differences between a forward vs. a backward chained read can be fairly
dramatic.

For instance I have spent the last 3+ years supporting a large AMISYS site.
As you maybe aware this software is a combination of Powerhouse, COBOL,
FORTRAN and Pascal.

Anyway one day the User community starting reporting broken chains in the
SERVICE data set. Knowing how the table is populated I felt that this was
very unlikely, but we had recently experienced a rash of power related
system crashes so I took a look at the situation.

In this case instead of using an Adager or other Third party solution I
using FORTRAN wrote my own verification routine. In fact I wrote 2, one for
a forward verification and the second for a backward verification.

To make this short, the backward chained read, which like it's brother
performed a serial read of the primary auto-master, did a DBFIND to the
SERVICE data set and then compared the # of successful DBGET's to the chain
rec. count from the DBFIND.

The backwrd read takes approximately twice as long as the forward read.
Which indicates to me that IMAGE 'prefetch' is somehow encumbered during a
backward read.

Thank you.

Thomas G. Cook




"Burgess Peter" <[log in to unmask]> wrote in message
news:92vj9c03g5@enews2.newsguy.com...
> Hi
>
> Does anyone have any experience of switching on the Image prefetch switch
on
> their dB's. I am thinking about doing this on some pretty active databases
> and was wondering if anyone had any bad experiences with this feature.
>
> Regards
>
> Peter Burgess
> Microwarehouse
> http://www.microwarehouse.co.uk
>
>

ATOM RSS1 RSS2