2.4.{19,20,20-ac2} with contest
-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1
Here's a set of contest benchmarks with the new version of contest for 2.4.x including 2.4.20-ac2:
no_load:
Kernel [runs] Time CPU% Loads LCPU% Ratio 2.4.19 3 76 96.1 0 0.0 1.00 2.4.20 3 77 94.8 0 0.0 1.00 2.4.20-ac2 2 76 96.1 0 0.0 1.00 2.5.58 2 79 96.2 0 0.0 1.00 2.5.58-mm1 4 79 96.2 0 0.0 1.00 cacherun: Kernel [runs] Time CPU% Loads LCPU% Ratio 2.4.19 3 74 100.0 0 0.0 0.97 2.4.20 3 74 98.6 0 0.0 0.96 2.4.20-ac2 2 74 98.6 0 0.0 0.97 2.5.58 2 76 100.0 0 0.0 0.96 2.5.58-mm1 4 77 97.4 0 0.0 0.97 process_load: Kernel [runs] Time CPU% Loads LCPU% Ratio 2.4.19 3 122 56.6 102 41.0 1.61 2.4.20 5 123 56.9 103 39.8 1.60 2.4.20-ac2 2 103 70.9 66 26.2 1.36 2.5.58 2 92 81.5 27 15.2 1.16 2.5.58-mm1 3 94 80.9 29 16.0 1.19 ctar_load: Kernel [runs] Time CPU% Loads LCPU% Ratio 2.4.19 3 96 81.2 1 4.2 1.26 2.4.20 3 97 81.4 1 4.1 1.26 2.4.20-ac2 4 108 71.3 1 5.6 1.42 2.5.58 3 117 82.1 1 6.0 1.48 2.5.58-mm1 3 109 81.7 1 4.6 1.38 xtar_load: Kernel [runs] Time CPU% Loads LCPU% Ratio 2.4.19 3 107 71.0 1 4.7 1.41 2.4.20 3 116 66.4 2 6.0 1.51 2.4.20-ac2 4 121 63.6 1 6.6 1.59 2.5.58 3 122 80.3 1 6.6 1.54 2.5.58-mm1 3 121 76.0 1 6.6 1.53 io_load: Kernel [runs] Time CPU% Loads LCPU% Ratio 2.4.19 3 286 26.6 23 15.3 3.76 2.4.20 3 219 34.2 17 16.0 2.84 2.4.20-ac2 4 240 30.8 17 15.0 3.16 2.5.58 3 153 54.9 8 14.3 1.94 2.5.58-mm1 6 156 51.3 9 14.7 1.97 read_load: Kernel [runs] Time CPU% Loads LCPU% Ratio 2.4.19 3 133 58.6 9 5.3 1.75 2.4.20 5 120 64.2 32 9.2 1.56 2.4.20-ac2 2 131 59.5 10 6.9 1.72 2.5.58 3 96 82.3 9 5.2 1.22 2.5.58-mm1 3 100 81.0 6 6.0 1.27 list_load: Kernel [runs] Time CPU% Loads LCPU% Ratio 2.4.19 3 88 86.4 0 5.7 1.16 2.4.20 3 88 85.2 0 8.0 1.14 2.4.20-ac2 1 89 84.3 0 7.9 1.17 2.5.58 3 91 85.7 0 8.8 1.15 2.5.58-mm1 2 92 83.7 0 9.8 1.16 mem_load: Kernel [runs] Time CPU% Loads LCPU% Ratio 2.4.19 3 96 78.1 43 1.0 1.26 2.4.20 3 95 80.0 45 1.1 1.23 2.4.20-ac2 3 101 77.2 91 4.0 1.33 2.5.58 3 107 73.8 66 1.9 1.35 2.5.58-mm1 3 104 75.0 50 1.0 1.32 dbench_load: Kernel [runs] Time CPU% Loads LCPU% Ratio 2.4.19 3 149 49.0 7 38.3 1.96 2.4.20 3 148 49.3 7 38.5 1.92 2.4.20-ac2 3 111 67.6 3 23.4 1.46 2.5.58 3 122 64.8 3 24.6 1.54 2.5.58-mm1 3 118 66.9 3 22.0 1.49
The different scheduler in ac2 shows up in process_load. The ide changes appear to make io loads tend towards 2.5 results in places. ac2 is also prone to the same out of memory unable to fork problem immediately after io_load which goes away if overcommit is enabled.
* Note to those who are not aware this is an issue and using contest: * Unfortunately by running make oldconfig && dep again I've changed the fs layout and will need to start with a new baseline from this point on. The number of page faults changes after this is done and kernel compilation takes a different length of time. Normally the number of page faults between runs is very close. And yes this does make a significant difference to the results.
Con
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.0 (GNU/Linux)
iD8DBQE+J3AaF6dfvkL3i1gRApwYAJ9AyYtgyAbzpbE2rviKOuQUAvOYKQCffgp0 8osik5cj300BDMLnj+lTwrU=
=xKSy
-----END PGP SIGNATURE-----