context-switch under 2.4.18pre7aa2 and 2.4.18pre7+O(1)J9

Duc Vianney (dvianney@us.ibm.com)
Fri, 1 Feb 2002 16:53:16 -0600


>hello, I'd be curious if you could add a column for 2.4.18pre7aa2 too, I
>did some optimization there, and I'd like to be sure it doesn't perform
>worse than 2.4.17 at least :). thanks,

>Andrea

The following results are context-switch latency time in microseconds,
taken under various test scenarios for lat_ctx.
Test configuration: smp 2-way 550 MHz, 512MB, smp kernel built.
Kernel under test: 2.4.17base, 2.4.17+O(1)J9, 2.4.18pre7base,
2.4.18pre7+O(1)J9, and 2.4.18pre7aa2.

Basic Results:
=> ctx in 2.4.18pre7aa2 is as fast as in 2.4.17 base.
=> ctx in 2.4.17pre7aa2 seems faster than in 2.4.18pre7 base.
=> ctx in 2.4.18pre7 base is slightly slower than in 2.4.17 base.
=> O(1)J9 helps ctx in either 2.4.17 base or 2.4.18pre7 base.

Test Data:

Test case: lat_ctx -s 0 2 4 8 16 32 64
Base O(1)J9 Base O(1)J9 aa2
# of processes 2.4.17 2.4.17 2.4.18pre7 2.4.18pre7 2.4.18pre7
2 3.03 0.97 3.27 0.93 3.07
4 3.55 4.97 3.80 4.28 3.51
8 4.41 4.43 4.55 3.92 4.39
16 4.78 3.82 4.71 3.50 5.35
32 7.29 4.56 5.70 4.53 6.62
64 8.37 5.62 7.73 5.73 7.49

Test case: lat_ctx -s 16 2 4 8 16 32 64
Base O(1)J9 Base O(1)J9 aa2
# of processes 2.4.17 2.4.17 2.4.18pre7 2.4.18pre7 2.4.18pre7
2 16.15 13.93 16.83 13.80 16.25
4 16.13 16.51 16.76 15.18 16.19
8 16.38 15.20 17.24 15.32 16.20
16 19.34 16.54 25.57 16.48 18.53
32 39.93 32.65 38.95 27.77 22.33
64 49.87 49.91 48.53 45.73 52.28

Test case: lat_ctx -s 32 2 4 8 16 32 64
Base O(1)J9 Base O(1)J9 aa2
# of processes 2.4.17 2.4.17 2.4.18pre7 2.4.18pre7 2.4.18pre7
2 24.73 24.29 25.36 23.12 24.68
4 24.85 25.22 25.37 25.30 25.08
8 30.94 25.04 33.86 24.61 31.34
16 74.32 48.32 58.81 43.20 35.71
32 94.41 89.62 77.29 93.76 79.53
64 96.91 94.29 100.88 92.95 102.02

Test case: lat_ctx -s 64 2 4 8 16 32 64
Base O(1)J9 Base O(1)J9 aa2
# of processes 2.4.17 2.4.17 2.4.18pre7 2.4.18pre7 2.4.18pre7
2 42.06 40.72 42.88 41.54 42.46
4 43.28 42.85 56.71 46.40 45.38
8 105.56 58.41 102.62 55.49 54.56
16 182.84 169.94 141.00 164.39 124.70
32 182.81 175.70 184.79 170.11 187.15
64 184.46 178.63 188.30 184.97 189.13

Duc.... dvianney@us.ibm.com

-
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/