HP3000-L Archives

May 1996, Week 3

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:
Gilles Schipper <[log in to unmask]>
Reply To:
Gilles Schipper <[log in to unmask]>
Date:
Thu, 16 May 1996 16:49:36 -0400
Content-Type:
text/plain
Parts/Attachments:
text/plain (39 lines)
I think the culprit here is likely that the automatic master containing 5.0
million records is beginning to experience the classic "clustering"
problem, whereby a single DBPUT to this master set (caused indirectly by a
DBPUT to a related detail) is resulting in a large number of physical I/O's
- which in turn is causing your system's performance to seriously degrade.
Actually the system's performance is exactly as it was, but the proportion
of its resouces being allocated to "useful" work is significantly lowered
due to the high volume of overhead I/O's being performed due to this
"clustering".
 
To determine whether or not this is really a clustering problem, you need to
run a program call HOWMESSY (from Robelle Consulting) or DBLOADNG (from the
Interex contributed library) against the database in question.
 
The output from these programs include a column entitled "max.blocks" or to
that effect. If this number is very high (say greater than 100) AND the
percentage of secondaries is greater than zero for the automatic master in
question, then clustering is likely the problem.
 
The solution is to increase the capacity of this automatic master to some
quite larger number (since this is an AUTOMATIC master, it should not occupy
that much extra disk space). The higher the number, the more likely your
chance of sucess - although it is entirely possible to make matters worse.
You may need to perform a little trial and error. Try one capacity, run
HOWMESSY, then try another capacity, etc. I realize that expanding a
5million entry master dataset will take quite a bit of time - but so it
goes. There is a technique I know of that can shortcut the trial-and-error
effort, but it can be quite risky if you don't know the intricacies of root
files and some tricks with the popular tools at hand (such as ADAGER).
 
I'm sure others will add their thoughts. If you need more specific
information, please ask.
-----------------------------------------------------------------------------
   Gilles Schipper                    Voice:      905/889-3000
   GSA Inc.                           Fax:        905/889-3001
   300 John Street, Box 87651         Internet:   [log in to unmask]
   Thornhill, ON  Canada  L3T 7R4     Compuserve: 71203,474
-----------------------------------------------------------------------------

ATOM RSS1 RSS2