[PATCH] powerpc: Fix database regression due to scheduler changes
Nick Piggin [Wed, 16 Nov 2005 12:37:53 +0000 (23:37 +1100)]
PowerPC's NUMA domain doesn't currently set up some of the newer
sched-domains parameters.

Brian Twichell <tbrian@us.ibm.com> discovered and diagnosed a 1.5% OLTP
database regression on a 4 core POWER5 system that was due to the use of
NUMA scheduling on ppc64.

This patch applies some saneish values to the parameters, in line with
other architectures. This solves the regression.

Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Paul Mackerras <paulus@samba.org>

include/asm-powerpc/topology.h

index 015d287..0e8c20c 100644 (file)
@@ -41,6 +41,10 @@ static inline int node_to_first_cpu(int node)
        .cache_hot_time         = (10*1000000),         \
        .cache_nice_tries       = 1,                    \
        .per_cpu_gain           = 100,                  \
+       .busy_idx               = 3,                    \
+       .idle_id                = 1,                    \
+       .newidle_idx            = 2,                    \
+       .wake_idx               = 1,                    \
        .flags                  = SD_LOAD_BALANCE       \
                                | SD_BALANCE_EXEC       \
                                | SD_BALANCE_NEWIDLE    \