Re: SMP and 7.x performance questions?

On Mar 8, 5:28pm, Fernando Fernandez wrote: > Subject: Re: SMP and 7.x performance questions? > Dave Kosenko wrote: > > A small SMP is a bit problemmatic, though. For some reason, it gets hard to > > load balance across only 2 cpus. For those in such a situation, I recommend > > trying to tun the system as though it were a uniprocessor (set the SINGLE_CPU > > parameter on). This introduces some optimizations through the code stream > > (avoids doing things that are only pertinant in a multi-proc environment) > > that can help. Once you are 4 procs or better, though, you are better off > > tuning for SMP. > > I already heard in several places that SMP will only work well up to 8 CPU. > If this theory is correct, we should only use SMP with 4 to 8 CPU's, then > start with MPP loosely-coupled architectures. This somewhat depends on the actual box. It varys between 8 and 32 I expect depending on manufacturer. > > Now back to Informix... This 7.1 performance problem in few CPU machines > raises another question: should we use disk fragmentation in those cases? > I believe this must be considered carefully because disk fragmentation is > strongly dependent on parallel execution. > > For example: I think that round robin fragmentation is not worth with low > number of CPU's, but hashing is a good policy because of the method of > skipping fragments. Round Robin is almost never a good idea as it really doesn't really provide benefits just a whole host of performance problems. Well thats my opinion anyway! But fragmentation isn't used only for parrallelism. You can use it to stop a query from accessing multiple fragments. This means that you can cut the number of index level searches you need to find a row. This works even on single cpu boxes. As the engine is threaded you should see benefit from fragmentation even on boxes with only 2 cpus if the query is complex enough. You might even see benefit on a single cpu box as the engine may use it's thread capability to switch fragment search threads when one becomes I/O blocked. Cheers - Jim -- ----------------------------------------------------------------------------- Jim Gordon DHL Airways Inc. jgordon@us.dhl.com ----------------------------------------------------------------------------- My opinions are my own. They may vary with time but they remain mine!