Oracle on VMS performance issue 
Author Message
 Oracle on VMS performance issue
Hi there!

We're running Oracle 7.3.2 on OpenVMS 7.1 (Alpha Server). The database
resides on a stripe-set. We suffer from performance problems related to
disk I/O. How does stripe size affect performance? I belive 64k stripes
are common with VMS. What would be the recommended stripe size, with
Oracle using 4k block-size? Does anyone have experience with this
configuration?

Thank you for your help!

Zeth Johansson



Mon, 12 Mar 2001 03:00:00 GMT
 Oracle on VMS performance issue

You don't want an Oracle block to span multiple stripes.  Your real problem
is how much IO your applications are performing.  I would start looking
at v$sqlarea and the number of IOs.  Run explain plans on the top
statements, and see if these can be tuned.

If you really want to brute force the application with hardware add more
memory cache at the disk or OS level.  Adding more buffer cache to
the sga may help, look at your buffer hit ratio.


Quote:
> Hi there!

> We're running Oracle 7.3.2 on OpenVMS 7.1 (Alpha Server). The database
> resides on a stripe-set. We suffer from performance problems related to
> disk I/O. How does stripe size affect performance? I belive 64k stripes
> are common with VMS. What would be the recommended stripe size, with
> Oracle using 4k block-size? Does anyone have experience with this
> configuration?

> Thank you for your help!

> Zeth Johansson



Wed, 14 Mar 2001 03:00:00 GMT
 Oracle on VMS performance issue

Quote:

>Hi there!

>We're running Oracle 7.3.2 on OpenVMS 7.1 (Alpha Server). The database
>resides on a stripe-set. We suffer from performance problems related to
>disk I/O. How does stripe size affect performance?

Mimimally so, unless you go for the extremes.
Hardware or software striping? Large or small db_block_size

You want to use UTLBSTAT/UTLESTAT or similar over a representative
test period to find whether you are doing a lot of sequential reads
or not (compare blocks_read with reads!).
If lotsa sequential reads then maybe up the read_ahaed multi block
count and use a smaller stripe size to ATTEMPT to get parallel IO in.

If mostly random IO, then just ose a largish stripe ( > 10 * block_size )
to avoid boundery fragmentation.

Quote:
> I belive 64k stripes are common with VMS.


Quote:
>What would be the recommended stripe size, with
>Oracle using 4k block-size?

64 Kb is probably fine ( 128 blocks) .

Quote:
> Does anyone have experience with this configuration?

Whether you can really eploit striping depends on the exact controllers,
and disks being used, the excess CPU power and spare IO bandwith.

hth,
        Hein.



Sat, 17 Mar 2001 03:00:00 GMT
 Oracle on VMS performance issue
just remmeber to NOT put your redologs and system tablespace
on the stripes disk. put them only on regular one.

Asaf.
============================================

Quote:

>You don't want an Oracle block to span multiple stripes.  Your real problem
>is how much IO your applications are performing.  I would start looking
>at v$sqlarea and the number of IOs.  Run explain plans on the top
>statements, and see if these can be tuned.

>If you really want to brute force the application with hardware add more
>memory cache at the disk or OS level.  Adding more buffer cache to
>the sga may help, look at your buffer hit ratio.




Quote:
>> Hi there!

>> We're running Oracle 7.3.2 on OpenVMS 7.1 (Alpha Server). The database
>> resides on a stripe-set. We suffer from performance problems related to
>> disk I/O. How does stripe size affect performance? I belive 64k stripes
>> are common with VMS. What would be the recommended stripe size, with
>> Oracle using 4k block-size? Does anyone have experience with this
>> configuration?

>> Thank you for your help!

>> Zeth Johansson



Sat, 17 Mar 2001 03:00:00 GMT
 Oracle on VMS performance issue

Quote:

> You don't want an Oracle block to span multiple stripes.  Your real problem
> is how much IO your applications are performing.  I would start looking
> at v$sqlarea and the number of IOs.  Run explain plans on the top
> statements, and see if these can be tuned.

> If you really want to brute force the application with hardware add more
> memory cache at the disk or OS level.  Adding more buffer cache to
> the sga may help, look at your buffer hit ratio.

More hardware cache probably won't help. You're describing 3 levels of
cache: the application (Oracle), the host system (VIOC?), and the hardware
(HS controller?). For the most part, unless outer cache layers are larger
than the inner layers, the contents of the outer cache will do nothing more
than duplicate the inner layer cache. But the outer cache will never be
"hit" because the request will be satisfied by the inner layer. It's mostly
a waste of $$$ and can actually slow performance.

One feature we're waiting for is supposed to be in the hs*80 controllers:
read ahead cache. For read transactions with locality, and especially for
sequential reads, this should help performance.

        Bob Kaplow      




Sun, 18 Mar 2001 03:00:00 GMT
 
 [ 5 post ] 

 Relevant Pages 

1. Compatibilty Issues between Alpha VMS 7.3 and Oracle 7.3.2

2. Compatibilty Issue between Alpha VMS 7.3 and Oracle 7.3.2

3. Compatibility Issues between Alpha VMS 7.3 and Oracle 7.3.2

4. Performance problems with Oracle 7.3.3.6 on Open VMS Alpha 7.1

5. Oracle/VMS performance problems with Multithreaded Server

6. Oracle VMS 7.1 performance

7. OPENROWSET performance issue or SET FMTONLY OFF issue

8. 23957-IL-Peoria-DEC-VMS-C-ORACLE-DEC VMS, C, Oracle

9. Performance issues with linked Oracle database

10. Oracle Linked Server Performance Issues

11. ENC: Oracle Performance Issue

12. Oracle visual basic performance issue.


 
Powered by phpBB® Forum Software