TM1 Benchmarks: Power7 AIX vs Nehalem Windows
Posted: Mon Jun 07, 2010 9:47 pm
After 9 months of external TM1/BI hosting, we’re planning to deploy a new TM1/BI architecture hosted internally. Poor Perspectives VPN performance and easier backend integration are the two key motivations behind the move. Additionally, we anticipate TM1’s usage continuing to grow as data sets are added, planning models are evolved, and other systems are migrated to TM1.
In short, we want to invest in the fastest available solution available today. Conveniently enough, IBM has made the case that the Power7 chip is minimally 2x the performance of Intel Nehalem and Nehalem-EX options on a single thread basis. What this actually means in real world TM1 usage is of course another mater.
We’re likely going to receive a loaner machine from IBM to run benchmarks (P750, 2 x 8 core P7 3.3GHz, 256GB ram). While we’ll of course benchmark our own logic and bottleneck points, I’m curious what people’s thoughts are about developing standardized TM1 benchmarks? What would be valuable (but still generic) performance measurements and how would you propose testing the measurement in a way that anyone with TM1 could benchmark?
Suggestions on large “stock†multidimensional public data sets that could used a common baseline on which various static and dynamic calculations could be tested? Or I suppose part of the benchmark setup be to populate a defined table with random values. Anyone up for helping to develop such a benchmark? I could see olapforums.com becoming the reference authority for this type of information.
Thoughts on this and/or TM1 on AIX in general?
Thanks,
-- John
In short, we want to invest in the fastest available solution available today. Conveniently enough, IBM has made the case that the Power7 chip is minimally 2x the performance of Intel Nehalem and Nehalem-EX options on a single thread basis. What this actually means in real world TM1 usage is of course another mater.
We’re likely going to receive a loaner machine from IBM to run benchmarks (P750, 2 x 8 core P7 3.3GHz, 256GB ram). While we’ll of course benchmark our own logic and bottleneck points, I’m curious what people’s thoughts are about developing standardized TM1 benchmarks? What would be valuable (but still generic) performance measurements and how would you propose testing the measurement in a way that anyone with TM1 could benchmark?
Suggestions on large “stock†multidimensional public data sets that could used a common baseline on which various static and dynamic calculations could be tested? Or I suppose part of the benchmark setup be to populate a defined table with random values. Anyone up for helping to develop such a benchmark? I could see olapforums.com becoming the reference authority for this type of information.
Thoughts on this and/or TM1 on AIX in general?
Thanks,
-- John