At 06:44 AM 11/24/2009, you wrote:
>2009/11/24 Johan De Meersman <vegivamp@stripped>:
> > If you are wondering about parallel query execution (that is, splitting a
> > single query over multiple cores for faster execution), that is currently
> > not supported by MySQL.
>Probably is something stupid, but could that be done with ndb cluster
>on a single host? Anyway, I suppose performance loses on distributed
>joins and so on would outcome multiple-core benefits. And for most
>queries, the bottleneck is usually on disk access, not processor. Has
>anybody done any serious testing on this?
Well it all depends on the SQL that is being executed, the table
structure and the size of the query. Now for a particular case you can do
your own benchmarking quite easily to see if disk speed is more relevant
than CPU speed. Copy your tables into a MEMORY table and do the joins
there. Compare that to a disk join (reset the query cache) and see the
improvement. I'm guessing you will probably see a 300% improvement over
disk. As mentioned earlier, MySQL does not scale up very well with multiple
processors which is why it is better to scale out horizontally with clusters.