R
Richard Banville
Guest
For data scan threads it is really trial and error as Paul already mentioned. Balancing CPU usage and I/O throughput is the key to getting this optimal. The idea of 1.5 threads * CPUs is to eliminate and wasted CPU during this part of the index rebuild. With today's machines having so many CPUs, this number can become ridiculous since the file system will become the bottleneck. Having too many threads that are not improving I/O rates will introduce contention/concurrency issues and may actually decrease performance. So you need enough datascan threads to use all the available CPU resources up to the point where I/O rates no longer improve. I know everyone wants a simple formula to just apply but there really isn't one that fits every deployment. The experience of others with success running this in similar deployments to yours is the best resource specific tuning suggestions.
Continue reading...
Continue reading...