Re: lies damn lies and benchmarks

in article YqPC8.3$_D5.11748_at_news.uswest.net, Pablo Sanchez at pablo_at_dev.null wrote on 5/10/02 5:55 AM:

> Are you the Mark Townsend of Oracle? Why not lobby your TPC-C
> representatives to make the motion at the -C meeting for the metric?
> I think it has a lot of merrit.

One of my more favourite topics, and as I sit next to him, he gets to hear my views almost daily :-)

I'd actually like to see a seperate HA benchmark set up, rather than combine it with the TPC-C. Idea would be to set a minimum level of TPS through which all backups need to be made, then introduce a series of common faults, and record the time/cost required to recover and get back to the prescribed throughput levels. Have two levels of faults in the BM - unplanned events (disk crashes etc), and a superset level also including planned events, such as user error, maintenance operations etc. Any remote mirroring etc would need to be able to keep up with the throughput, survive network outages etc.

There's already been some work done in academia on identifying the faults and building a repeatable framework for fault injection, so it wouldn't be a big step to turn this into a benchmark.It would play well with some of the Oracle features as well - fast recovery, online ops, failover to warm caches, standby, HARD initiative, certified configs etc.

A very useful exercise, and one that would help the industry in a large way - imagine if the hardware vendors and software vendors actually worked together to solve and standardize operations around this little problem.

Of course, these are my own views, and not necessarily those of Oracle's. Received on Fri May 10 2002 - 09:51:11 CDT