On Fri, 13 Feb 2004 11:34:34 -0600 TriKster Abacus <[EMAIL PROTECTED]> wrote:
[ lots of stuff that we don't need to repeat snipped ] TriKster, The essential flaw in your "benchmarks" is this: all of these items are single activities, whereas the changes in kernel 2.6 are not designed to improve the operation of single activities, but rather to improve the aggregate thruput of the system for l-o-t-s of activities, i.e. 1) more robust scalibility for >4 processors 2) better response for interactive jobs (i.e. no job should starve for service while a cpu hog is getting its cut off the top). This means, inevitably, that certain of your loved ones (DVD ripping, for example) may not be able to monopolize the CPU and thus may exhibit extended run times. If you want a fair picture of 2.6 vs. 2.4 operations, you must construct a very mixed, stressful combination of tasks to run opposite X sessions and measure not only the job throughput but the response time effect on the X sessions. Then you would repeat the study varying the number of cpus, memory, etc. I'm pretty sure you will find the same results that the IBM 8-way processor benchmarks did - individual component run times may lengthen, but the system is capable of accomplishing more concurrent work. Also, the IBM benchmarks demonstrated that 2.6 consumes a lot less kernel mode and a lot more user mode CPU time while getting more work through the box. I, personally, don't need to do this sort of benchmarking, because I remember full well the effect of running major compiles on 2.4 while trying to do something else useful. It wasn't a pretty picture unless you bumped the nice values for X, and it now works quite well without screwing with any settings, so that's as accurate a report as I need. YMMV. Enjoy, -- Collins - Denver Area - Gentoo stable kernel 2.6.2-rc1 -- [EMAIL PROTECTED] mailing list
