aboutsummaryrefslogtreecommitdiffstats
path: root/Documentation/kernel-parameters.txt
diff options
context:
space:
mode:
authorIngo Molnar <mingo@elte.hu>2007-07-09 12:51:57 -0400
committerIngo Molnar <mingo@elte.hu>2007-07-09 12:51:57 -0400
commit0437e109e1841607f2988891eaa36c531c6aa6ac (patch)
treee9d8f170786f7e33d4c5829cb008cf38d42a2014 /Documentation/kernel-parameters.txt
parent0e6aca43e08a62a48d6770e9a159dbec167bf4c6 (diff)
sched: zap the migration init / cache-hot balancing code
the SMP load-balancer uses the boot-time migration-cost estimation code to attempt to improve the quality of balancing. The reason for this code is that the discrete priority queues do not preserve the order of scheduling accurately, so the load-balancer skips tasks that were running on a CPU 'recently'. this code is fundamental fragile: the boot-time migration cost detector doesnt really work on systems that had large L3 caches, it caused boot delays on large systems and the whole cache-hot concept made the balancing code pretty undeterministic as well. (and hey, i wrote most of it, so i can say it out loud that it sucks ;-) under CFS the same purpose of cache affinity can be achieved without any special cache-hot special-case: tasks are sorted in the 'timeline' tree and the SMP balancer picks tasks from the left side of the tree, thus the most cache-cold task is balanced automatically. Signed-off-by: Ingo Molnar <mingo@elte.hu>
Diffstat (limited to 'Documentation/kernel-parameters.txt')
-rw-r--r--Documentation/kernel-parameters.txt43
1 files changed, 0 insertions, 43 deletions
diff --git a/Documentation/kernel-parameters.txt b/Documentation/kernel-parameters.txt
index af50f9bbe68e..4d880b3d1f35 100644
--- a/Documentation/kernel-parameters.txt
+++ b/Documentation/kernel-parameters.txt
@@ -1014,49 +1014,6 @@ and is between 256 and 4096 characters. It is defined in the file
1014 1014
1015 mga= [HW,DRM] 1015 mga= [HW,DRM]
1016 1016
1017 migration_cost=
1018 [KNL,SMP] debug: override scheduler migration costs
1019 Format: <level-1-usecs>,<level-2-usecs>,...
1020 This debugging option can be used to override the
1021 default scheduler migration cost matrix. The numbers
1022 are indexed by 'CPU domain distance'.
1023 E.g. migration_cost=1000,2000,3000 on an SMT NUMA
1024 box will set up an intra-core migration cost of
1025 1 msec, an inter-core migration cost of 2 msecs,
1026 and an inter-node migration cost of 3 msecs.
1027
1028 WARNING: using the wrong values here can break
1029 scheduler performance, so it's only for scheduler
1030 development purposes, not production environments.
1031
1032 migration_debug=
1033 [KNL,SMP] migration cost auto-detect verbosity
1034 Format=<0|1|2>
1035 If a system's migration matrix reported at bootup
1036 seems erroneous then this option can be used to
1037 increase verbosity of the detection process.
1038 We default to 0 (no extra messages), 1 will print
1039 some more information, and 2 will be really
1040 verbose (probably only useful if you also have a
1041 serial console attached to the system).
1042
1043 migration_factor=
1044 [KNL,SMP] multiply/divide migration costs by a factor
1045 Format=<percent>
1046 This debug option can be used to proportionally
1047 increase or decrease the auto-detected migration
1048 costs for all entries of the migration matrix.
1049 E.g. migration_factor=150 will increase migration
1050 costs by 50%. (and thus the scheduler will be less
1051 eager migrating cache-hot tasks)
1052 migration_factor=80 will decrease migration costs
1053 by 20%. (thus the scheduler will be more eager to
1054 migrate tasks)
1055
1056 WARNING: using the wrong values here can break
1057 scheduler performance, so it's only for scheduler
1058 development purposes, not production environments.
1059
1060 mousedev.tap_time= 1017 mousedev.tap_time=
1061 [MOUSE] Maximum time between finger touching and 1018 [MOUSE] Maximum time between finger touching and
1062 leaving touchpad surface for touch to be considered 1019 leaving touchpad surface for touch to be considered