Oracle + SAP updates (long post)

[Dan Hetzel]

Hi all,

One of my clients presently uses the following architecture:

3xD380 servers oracle 7.3.4 - production
1xD380 server oracle 7.3.4 - dev
1xD380 server SAP 4.0.b - production
1xD380 server SAP 4.0.b - dev
1xD380 server legato 5.5.1
+ a couple of test systems...
2xEMC disk arrays

All servers running HP-UX 10.20

In the target configuration, all servers will be replaced by L2000 systems used for:

3 servers Oracle 8.0.6 production (cluster)
1 server Oracle 8.0.6 dev
1 APP server SAP 4.6.c prod (cluster with DB server)
1 DB server SAP 4.6.c prod (cluster with APP server)
1 server SAP (APP+DB) dev
1 cluster (2 servers) with LEGATO, OpenView and autosys
2 EMC disk arrays (new models)
1 Connectrix SAN

Everything (hard/soft/disks/network) is changing so the upgrade has a good chance to be quite tricky.

Considering that they will have all the new equipment added to the existing servers I was wondering which of the following upgrade paths was making sense:

Option 1:
Install all new platforms, O/S, patches, software (oracle/SAP/legato/OpenView/ServiceGuard/...),
SAN, EMC... in parrallel with the present production network.
Once all tuning and testing is done, transfer data and go live ??

Option 2:
Replacing old systems one at a time.
Swap EMC disk arrays at the end (with data transfer)

I would personally prefer option 1, as the production will be untouched until the very last step, but this one will be kind of a 'giant leap'.

How would you personally deal with this?

I tried to make it as short as possible but if had enough courage to read until here, you already deserve some points ;-)

Thanks,

Dan

[John Palmer · Accepted solution]

Hi Dan,

I'd vote strongly for option 1 provided you have a long enough time window to actually transfer the data at 'go live' time.

Providing your testing is comprehensive, it won't be such a 'giant leap' will it? and if things do go 'pear shaped' you would be able to revert to the old servers.

Regards,
John

[James R. Ferguson]

Hi Dan:

Advice is always free! ;-)

My choice would be option-1. Implementing and testing MC/ServiceGuard is no small endeavor. By deferring the cutover until everything but the data is ready to "swap" means that you can really test everything well. At that point, "all" that's left to do is re-point data. Good Luck!

...JRF...

[Dan Hetzel]

Thanks guys !

I know that this looks like a HUGE task but my client already asked for HP support.
HP will install and test MCServiceGuard with the SAP smart plugin as well as OpenView.

I'm still wondering what would be the best strategy to transfer data between the old and new EMC.
I've been told by EMC tech that we couldn't use SRDF because of "discrepancy in O/S" (10.20/11.0). I'm not sure it makes sense...

If we want to be able to test everything on the new platforms, we'll have to use the production data (about 40Gb without the indexes). Exporting/transfer/importing the data will be quite a lenghty process, I'm afraid.

How would you proceed ?

Dan

[Dan Hetzel]

Oops !! a typo: 400Gb in 6 oracle instances

Dan

[Dragan Krnic]

a 100 GB per server. Not exactly a piece of cake.

I would export each database as fast and possible and start archiving from then on on the source server. Redoing the archived logs at the target server should keep the databases up to date or almost. The "great leap" is then just another daily routine.

[CHRIS ANORUO]

Don't use the default buffer in your export scripts. I have modified my ora_load and ora_dump scripts with buffer of 5mb for faster dumps. Make sure you have a good kernel the does dynamic buffering.

[Dan Hetzel]

Thanks John, James, Dragan and Chris.

Your answers confirmed that this was the only way to go. The most difficult point for me now, will be to find the right arguments to force the dba and SAP admin to accept that they will have to administer a dual environment for quite a while...

Any good arguments in mind ? ;-)

Best regards,

Dan