Multiprocessor alternance
Hihoo! I'm very surprised, as i've compiled kernels for a P4 hyperthreading and for a true P2 multiprocessor, and, watching peacefully gkrellm monitor, i've seen that cpu's were working in alternance. Ex.: one at 80% and he other at 20%, or one at 99% and the other at 1%, and never both at 100%. So, as i can't sleep and eat since i've seen that, my only hope to understand this is you. Please, help me.
|
hyperthreading
Hello
My experiance with the 2.4 series kernels is that cpu's don't share work loads evenly between them. I have heard that the 2.6 series kernels do share the load better. On a dual hyperthreaded xeon system, I have only seen maximum use of all cpu's when running 4 seti@home processes at the same time. For routine daily activities like compiling kernels :) there is always a cpu or two that make nothing but heat. |
Thank you. Any other experiences?
|
My (hyperthreaded) dual xeon box shows the same alternation.
When using 3 of the 4 available cpu's (seti@home), 3 out of the 4 cpu's are at 99-100% (most of the time). But, never the same 3 cpu's. Example, running only seti@home: Cpu0 : 0.0% user, 0.0% system, 100.0% nice, 0.0% idle Cpu1 : 0.0% user, 0.0% system, 0.0% nice, 100.0% idle Cpu2 : 0.0% user, 0.7% system, 99.3% nice, 0.0% idle Cpu3 : 0.0% user, 0.7% system, 99.3% nice, 0.0% idle And a few cycles later: Cpu0 : 0.0% user, 0.0% system, 0.0% nice, 100.0% idle Cpu1 : 0.0% user, 0.7% system, 99.3% nice, 0.0% idle Cpu2 : 0.0% user, 0.3% system, 99.7% nice, 0.0% idle Cpu3 : 0.0% user, 0.8% system, 99.2% nice, 0.0% idle I did experiment with disabeling hyperthreading (2 'real' cpu's instead of 4 'virtual'). The overall performance is better with 4 cpu's (although not that much). Did not try the 2.6 kernel yet, so cannot tell anything about that. |
Thanks again.
|
yes 2.4 kernels suck big time at multiprocessor balancing
definitely use the latest 2.6 as all modern distros now are 2.6 ready |
All times are GMT -5. The time now is 09:07 AM. |