Re: MVS and large database

In article <31E676C4.55D1_at_nla.gov.au>, Bruce Pihlamae <bpihlama_at_nla.gov.au> writes
>Brian McCabe wrote:
>> The site I am working at is also experiencing problems with Oracle
>> performance on MVS. We are planning to migrate from AIX to MVS and
>> although the database only currently stands at less than 100Gb, we
>> intend to reach pretty near to the 1Tb figure next year.
>>
 

>> Initial performance of Oracle on MVS is very disappointing and no
>> obvious bottle-neck has been determined to date. The Orace Instance is
>> idling, the network is not under any great stress and the MVS-Gurus tell
>> us that nothing is pushing the operating system or disk IO sub-system.
>>
>> A question for Bill - is there a repository of information on tuning
>> Oracle for MVS anywhere which contains information which is not in the
>> Oracle/MVS SYS Admin Guides? We have followed these instructions to the
>> letter and performance is still at least three times as poor as on the
>> AIX Box.
>>
>> Can anyone else offer any pointers in the right direction????
>
>Please excuse my ignorance but...
>
>If the cpu and i/o subsystem are underloaded then you might be queueing on
>something else either
>at the o/s, network, or database levels. I haven't done any MVS work for a
>while but can you set
>the number of worker tasks (in unix they're shadow processes that access the
>database for the
>client) higher so that more connections are active concurrently. Have you
>confirmed that there is
>no internal database lock contention. Are all your interactive and batch
>processes running at the
>same cpu priority. Is your database SGA too large and being paged out of real
>memory.
>
>--
> -----------------
> T T T T T T T T Bruce Pihlamae
> I I I I I I I I bpihlama_at_nla.gov.au
> I I I I I I I I National Library of Australia
> T T T T T T T T Phone: +616 262-1575
> ----------------- Fax: +616 273-2116
> ===================
> "If you swallow a live frog first thing in the morning;
> nothing worse will happen to either of you that day."

Bruce - thanks for the suggestions but I'm afraid we have investiagted all these! We have attempted to initiate different numbers of worker tasks via the DYNWORK MPM parameter (it defaults to approx 32000 but setting it lower will apparently give you better performance in some circumstances). We have also checked for contention at a database level and tried SGA's ranging in size from 10Mb upto 72Mb with no noticeable changes in performance. There is no paging at the Operating System level and no waits or excessive reloads/misses of SGA objects. Cache Hit Ratio is high and the network/CPU are both idling. There is a blockage somewhere but we ae really struggling to find it! Any other suggestions are more than welcome. Has anyone else out there experienced ether good or bad performance of reasonably-sized Oracle database on MVS running across SQLNET2 TCP/IP from non-MVS clients???

-- 
Brian McCabe
McCabe IT Consulting Limited
brian_at_mitc.demon.co.uk