VA7400 - write back cache

[Simon White]

Configuration is: Two Windows 2000 Adv. Servers (SP3) running in a cluster. Each server has two redundant paths to a VA7400(dual controller) by using two D8602B HBAs per server and two Brocade switches. Both servers are running Autopath 1.04 (No load balancing) and the driver version for the HBAs is v2.0.25.44. Firmware of VA is HP16, upgrading to HP18 in the next few weeks.

Our application is based on a Progress database and recently we have been experiencing corrupt data. The developers are telling us this is because we are using write back cache and they would like it turned off or switched to write through cache.

My understanding is that the VA7400 uses write back cache and this can not be changed.
Am I correct?


Also how does this relate to the disk properties from within W2K? If you look at the properties of a disk and then click hardware, all the LUNS are displayed and they all have an option to turn on/off "write cache enabled", by default it is on.
What would happen if I unchecked this box? Is this just an OS thing or will it affect the VA

Thanks
Simon

[Eugeny Brychkov]

Simon,
this should not cause data loss unless VA has dead completely losing its cache contents.
Can you please check or gather 'armdsp -a' and 'logprn -t All -v -a', zip them and attach to your next reply?
Did you see any warnings displayed by VA?
Please note that latest Autopath is 2.0. You should upgrade it also
Eugeny

[Vincent Fleming · Accepted solution]

Simon,

Your DBAs are misinformed (as you may have guessed) about write-back cache. In the olden days, yes, there were some problems with databases and write-back cache, but that problem was long ago fixed.

I'm told that the cache setting on the VA is meaningless - it's there for compatibility, and actually does nothing. In fact, there are several settings such as this. In the case of the cache, it's emulating a disk drive with write-back cache capability. The OS likes to be able to "turn it off"... even if turning it off does nothing.

My suggestion is to upgrade the FW asap. I'm not aware of any data corruption problems with HP16, but that means little.

Good luck,

Vince

[Roger_22]

The application developers are on drugs..... it's not the write back cache. The array may have a problem, but it's not the cache.

But, if you want to turn it off you can. Secure Mode. The controller ignore all the cache settings, but Secure Mode will disable write cache.

Upgrade your firmware to HP18, and upgrade your application developers.

r

[Simon White]

Thanks for the replies. This corruption is very intermittent and occurs when certain batch jobs are run over night. The problem I have is that copies of the database have been restored to other systems with internal storage and the errors in the data do not occur. My job is to prove that the VA is not the cause of this problem.
There are no errors on the VA or in any it's log files and personally I think it is the app that has a bug.

If I was to put the VA in Secure Mode, this would reduce performance dramatically, wouldn't it?

Simon

[Don Capps]

You might wish to try running Iozone. (http://www.iozone.org) and using the -+d option. This will tell Iozone to perform a diagnostic with byte level validation of all data written and read back. This is a very handy way of telling if there is something broken in the system that is causing corruption.