Re: [Q] Increase SGA, increase performance ???

On 06/06/1996 5:10PM, in message <dsmith.129.000F2EC4_at_gulfaero.com>, Doug Smith <dsmith_at_gulfaero.com> wrote:

> In article <4p6vgn$3am_at_news2.cais.com> duffy_at_cais.cais.com (MCC) writes:
> >From: duffy_at_cais.cais.com (MCC)
> >Subject: [Q] Increase SGA, increase performance ???
> >Date: 6 Jun 1996 16:03:35 GMT
 

> >We have SUN SParc 1000E (SOlaris 2.5) with Oracle 7.2.3 installed. This
> >SUN UNIX computer have 4 CPU and 128 MB RAM (Total). The SGA currently
> >setup size "8386600" and the database size is 1.5GB (total datafiles). I
> >don't know the SGA size is big enough or not? If I increase SGA size,
> >will it increase performance? How big SGA size is good value? Do I need
> >put more RAM or not? Thank you for help.
>
>
> I don't work with Sun's just hp and ibm, however, my workstation has 128meg
> ram! It's hard to imagine why you would have 4 cpu's with only 128 meg ram?
ru kidding ? What a statement ! It's like saying: " It's no point driving a car with a v8 when you only have a 60-liters tank.
>
> About the SGA, you have to look at performance numbers to determine sizes.
> Yes, if a buffer increase (increases the SGA) reduces the number of physical
> I/O's, then it will increase performance.
Yes, but there is no way around writing ALL deletes & updates to disk, in any way you implement the database. The bottle-neck often isn't memory, but the completely redundant & ad-hoc database-updates. WHen you minimize the number of editable tables your disk-IO overhead will dramatically decrease. ALso check your indexes on updateable tables, these often slow down transaction-management, since for every modified/deleted row the index has to be updated which can be very time-consuming! The 128 can then be used for read-only tables (system-tables, constants etc.) which help performance. There is no point in keeping large bulks of ad-hoc tabeldata in memory.
>
> A fairly easy way to experement is to change the setings in the init?.ora
> file to the oracle suggested small, med, large settings, then measure your
> performance. They are just guesses, but that's all you can do anyway without
 Yeah right, they are more like wild stabs in the dark.
> real performance analysis.

?????? How (un)real can it get ?

Hope this helps

l8er,
Edwin van der Sanden

biged_at_worldaccess.nl Received on Sat Jun 15 1996 - 00:00:00 CEST