As discussed on -hackers:

Patch to test performance for lwlock structure padding for 8.1beta

This patch is only expected to show improvement on larger SMP systems
where there is considerable lwlock activity, on related lock types
e.g. multiple backends doing tight nested loop joins at the same time as
lots of cache misses

Cost: 6x 100 bytes of memory on to each backend

Based on recommendations from CPU hardware optimization guides and one
previously successful test of DBT-2/TPC-C.

For testing only, at this stage.

Best Regards, Simon Riggs


---------------------------(end of broadcast)---------------------------
TIP 5: don't forget to increase your free space map settings