Con Kolivas recently posted some interesting kernel benchmarks generated with his contest benchmarking tool. Using resources provided by the OSDL, Con compared 2.4.19, 2.4.20, 2.5.49 and 2.5.50, each with a single processor enabled and with dual processors enabled. On KernelTrap you will find many of the tests, Con makes some quick observations about the results.
Linux: Benchmarking 2.4 vs 2.5, UP vs SMP
About The Author
Eugenia Loli
Ex-programmer, ex-editor in chief at OSNews.com, now a visual artist/filmmaker.
Follow me on Twitter @EugeniaLoli
can someone tell me what these numbers mean?
Read it, he explains it in there.
Can processor power really make all that much difference in I/O? Double digit second differences (between smp and up) are not easy to make. I know that on all my machines (running openbsd or mac osx depending on arch) flooring disk access doesn’t consume more than 50% of the processor. Can anybody comment on the quality of the I/O code with respect to CPU use in the linux kernel, or is this in the scheduler?
as per the topic, without any kind of explanation greater than a single sentence, it makes for poor reading….
shame kernel trap posted what looks like a draft article…
“Can processor power really make all that much difference in I/O? Double digit second differences (between smp and up) are not easy to make. I know that on all my machines (running openbsd or mac osx depending on arch) flooring disk access doesn’t consume more than 50% of the processor. Can anybody comment on the quality of the I/O code with respect to CPU use in the linux kernel, or is this in the scheduler?”
It’s not processor power. It’s concurrency. It’s not quality of I/O code. It’s quality of scheduling I/O. Allowing multiple I/O transactions to begin concurrently overlaps their latency.
that 2.4.x is actually faster than 2.5.x?
Will someone please explain (and thus making this an article)