Database mirroring, again
2006-02-08 00:07:03
Thank you all for your comments. This has been an interesting discussion
to follow.
My original incentive to change was performance. We're in a database I/O
bottleneck situation, and hardware mirroring/striping should definitely
help. I'm in the process of changing to it right now, and I'll report on
the actual performance change when I'm done.
Leigh: I am not using a battery-backed cache in the RAID controller. I
have some fear that it could actually make corruption more likely in a
hard crash situation. Better to allow the incomplete uncomitted
transaction to roll back.
We had previously changed to MIRRORWRITE DB PARALLEL to help performance
- with a database shadow table. It did help. I assume that the shadow
table will now be moot, and will not be used, once I finish undefining
all the TSM-created mirror copies.
Though I fear that this will be a slightly riskier configuration, I want
very much to measure the performance difference, to see if the risk is
worth it. It very well might be. There's only one way to find this out
on a really big TSM database, and that is to try it. It's going to take
another day to get there.
Again, thanks to all for your input on a basic topic that matters.
Roger Deschner University of Illinois at Chicago rogerd AT uic DOT edu
"Panel on shouting-match TV shows ends in shouting match"
--headline, Chicago Tribune, February 15, 2005
|
|
|