On 7/20/07, James Melin <[EMAIL PROTECTED]> wrote: > It means that IBM has z9 BC vs z900 performance benchmarks. MHZ ratings are > not meaningful across processor architetures, and are not management > meaningful to any useful degree. IBM's benchmarks of performance between a > z9 BC and a z900 are well understood by IBM and that compartive > information is management meaningful.
So I guess "management meaningful" means we know it's BS but the boss will buy it and it gets him off your back? Yes, sometimes you give in - like I did in the following anecdote :-) A large outsourcing company was using the IBM published ratings for various CPU models to normalize the CPU usage of each image. Simple - you divide the total cost of the computing center by the "normalized used MIPS" to give you the $ per normalized CPU hour, and then cut the bill for each LPAR. As you should know, the models are rated on "single z/OS image throughput" and MP overhead in MVS means a 20-way z/OS image does not do 10 times more than a 2-way image. The bean counters now conclude that normalization of the 20-way is different from the 2-way, so a single CPU hours on a 20-way will be cheaper than on a 2-way... Among others, we ran a production LPAR and a small test LPAR on two different machines: a 6-way and a 16-way. By swapping the two LPARs so that production runs on the 16-way, we saved money... Both were z990 so we used the same amount of CPU hours. But since the bean counters had decided the per-CPU capacity of a 16-way was only 75% of the CPU in a 6-way, we reduce our CPU cost with 25%. The test system did not use a lot, so we could afford to run that on a more expensive CPU. I have also defended this claiming the tests would be better on an expensive CPU :-) Rob ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [EMAIL PROTECTED] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390
