IDS and Sun

Has anyone any experience of running IDS 9.40 or 10 on Sun's new T1 processors (or should I say "low-heat, highly parallelized T1 processors"?) Thanks -- Neil Truby t:01932 724027 Director m:07798 811708 Ardenta Limited e:neil.truby@ardenta.com

No but do you have one going spare I can play on? I would be aware of the limitations of Hyperthreaded Processors and IDS that I vaguely remember someone posting here. Database are about data movement (including into/out of the CPU package). Perhaps ensuring CPU VPS are spread across the CPU packages by using the AFF_NPROCS and AFF_SPROC parameters would be useful to not bottleneck on data movement into and out of the same 'CPU package' may help? I'm guess this is kinda like a NUMA architecture in terms of bottlenecks? It can be better for avoid bottlenecks on disk controllers to not have too many active disks per controller so perhaps too many cores per CPU package actively doing data transfers may eventually bottleneck! This would need rethinking the AFF_ parameters since I am not sure that cores are numbered round robin across the CPU packages? Talk to IBM and get back to us please? Can you benchmark big parallel sorts using multiple CPU VPS using cores on the same CPU package vs multiple CPU VPS on cores on seperate CPU packages? You would need to manually bind processes to cores or even better like Art Kagel does with his NUMA stuff, CPU VPS to CPU packages rather than cores. Anyone know if pbind has been extended for this?

> Talk to IBM and get back to us please? Ah, you make it sound so easy ....