MOLNAR Ingo wrote:
> On Mon, 3 Aug 1998, Andre Derrick Balsa wrote:
> > Conclusion: if you want faster compilations, get a faster CPU. Going
> > from Linux 2.0.x to 2.1.x will get you a negligible performance gain for
> > this kind of task.
> this is not necessarily true for all systems. Although the system in
> question was a UP box, it's i think worth to mention that compilation
> speeds on SMP improve measurably with 2.1.
Because of finer-grained kernel locking, yes. Could you come up with
actual figures? I have some old numbers (2.1.6x, I think) that show the
improvement is marginal relative to 2.0.x.
And even so, kernel compilation is a _bad_ benchmark for SMP efficiency
improvements in 2.1.x vs. 2.0.x. How do you sort out the SMP performance
gains from the dcache performance gains? Tough.
> also, compilation _can_ be 'kernel-bound' on UP too, just try compiling
> Linux on a say 4MB box. (or 8MB)
Bad example: then you are not kernel bound, but memory (swap) bound. And
2.0.x is noticeably *faster* than 2.1.x in low memory machines.
I am repeating myself, but getting correct benchmarking numbers means
one must first identify the performance bottleneck, and then devise or
use the available tools to measure the actual bottleneck. Otherwise we
begin mixing variables, and the numbers loose any meaning.
Perhaps I should have been more explicit in my previous post, so:
*provided one has enough memory to avoid swapping*, kernel compilation
is basically CPU-bound (integer and cache/memory bandwidth).
In benchmarking, the devil is in the details. :-)
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to email@example.com Please read the FAQ at http://www.altern.org/andrebalsa/doc/lkml-faq.html