Re: SMP and 7.x performance questions?

On Mar 12, 11:54am, Fernando Fernandez wrote: > Subject: Re: SMP and 7.x performance questions? > Jim Gordon wrote: > > 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! > > Sorry Jim, but I don't agree with you. I believe Round Robin is a good > solution if you don't know how data is strutured *AND* you use several > processors *AND* use PDQ. > > Simply put: several processors accessing simultaneously several independent > disks has got to be faster (once again, assuming that you don't know how data > is strutured and you couldn't use hashing). > > Cheers, > Fernando >-- End of excerpt from Fernando Fernandez For those of you who are wondering what the start of this thread was, you should know that Fernando is responding to something I wrote between 3 and 6 months ago, maybe longer! Unfortunately I can't completely remember what I was talking about either! There was another comment to my remark at the time with which I agreed but that has slipped my mind. In this case I would just say that assuming plenty of processers and PDQ set up you can get good performance out of round robin. But a good dba should normally be able to find a fragmentation policy that provides at least that level of performance with general queries + a higher level of performance for certain classes of queries where the fragmentation policy can be used to reduce the amount of data searching required. This also leads to less load on the box allowing for additional unrelated work to be accomodated before needing a box upgrade. This should be true even in a decision support environment where query structure is admittedly more random. But a review of the queries used should give a dba some pattern to work with. Of course this suggests that you might start out with round robin until you get that info and that you might vary the fragmentation policy over time. I would suggest that it would be a rare database where a pattern of query use could not be determined over time. Consider also that the choice of fragmentation policy may not necessarily be directed at the biggest and slowest query pattern. Here the ability to do parrallel searches across multiple disks may provide better performance than a fragmentation policy that puts all the data for those queries on one drive. But there are probably secondary query patterns looking at much smaller amounts of data that would would seriously benefit from a fragmentation policy directed at them while still leaving a generally random distribution of data for the large queries. All of the above pre-supposes that yon management is willing to invest the time in selecting these strategies. 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!