>math. i'll post something about this soon.
Cool! thanks.
>i looked at doug's patch too, and it changes the "page_shift" value
>depending on the size of the hash table. again, this *may* cause unwanted
>interactions making the hash function degenerate for certain table sizes.
Agreed.
>ref: a stock 2.2.5 kernel
>
>p-al: a stock 2.2.5 kernel with your page struct alignment patch applied
>
>irq: a stock 2.2.5 kernel with your irq alignment patch applied
>
>both: a stock 2.2.5 kernel with both patches applied
*snip*
>ref: 4176.4 (s=27.45)
>
>p-al: 4207.9 (s=8.1)
^^^ it made _difference_
>
>irq: 4228.8 (s=11.70)
>
>both: 4207.9 (s=13.34)
^^^^^ strange...
>the irq patch is a clear win over the reference kernel: it shows a
Good ;)
>consistent 1.25% improvement in overall throughput, and the performance
>difference is more than a standard deviation. also, the variance appears
>to be less with the irq kernel. i would bet on a more I/O bound load the
>improvement would be even more stark.
>
>i'm not certain why the combination kernel performance was worse than the
>irq-only kernel.
Hmm I'll think about that...
>"Lynch" is a PhD thesis available in postscript at Stanford's web site for
>free. it's a study of different coloring methodologies, so it's fairly
>broad.
Thanks!! I'll search for it soon.
Andrea Arcangeli
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/