Well on high load, you shouldn't have an idle cpu anyway, so you would never
pass the requirements for the agressive -idle- steal even if it was turned
on. On low loads on HT, without this agressive balance on cpu bound tasks,
you will always load up one core before using any of the others.
For a brief period of time. But with active_load_balance() being called
on the idle processors, they will steal from a core/sibling pair to give
themselves something to do in, I should think, relatively short order.
Myself, I'm getting odd results on kernbench (kernel compiles). On a
4-proc + 4-sibling HT machine, I'm seeing:
2.5.68 HW HT on, regular scheduler
2.5.68-ht2 HW HT on, A9 hyperthreading scheduler siblings=2
The HT scheduler seems to give us wins in every category but elapsed time.
This was *with* the aggressive steal, so I've a bit more testing to try
without it, collecting more information to identify why elapsed time is
not dropping too.
Rick
make -j2
User System Elapsed %CPU
2.5.68 471.19 34.14 263.25 191%
2.5.68-ht2 335.54 24.46 257.86 139%
make -j4
User System Elapsed %CPU
2.5.68 581.93 40.37 164.36 378%
2.5.68-ht2 421.77 28.64 165.06 272%
make -j8
User System Elapsed %CPU
2.5.68 946.24 60.05 138.13 728%
2.5.68-ht2 685.05 43.38 138.59 525%
make -j16
User System Elapsed %CPU
2.5.68 954.35 61.12 139.17 729%
2.5.68-ht2 690.11 43.91 138.52 529%
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
This archive was generated by hypermail 2b29 : Wed Apr 30 2003 - 22:00:12 EST