Hello everyone,
I consider using two instances of the PG server on one node. The node
has two processors and each processor has 64GB memory assigned to it.
I'd like to
* restrict instance one to processor 0 and its memory and
* instance two to processor 1 and its memory
using numactl (or similar commands regarding cpuset) and partitioning.
I expect >30% speedup on some queries, because collecting data can be
done in parallel.
Some background information on the system: It's a regular 2 Xeon (each
4x2 cores due to hyperthreading) system with a lot of memory. Memory
assigned to the processor can be accessed quickly, memory assigned to
the OTHER processor is slow.
Does anyone here know a way to accomplish that? I think openstreetmap
map of germany (4GB) is a nice opportunity to test performance.
Looking forward to constructive answers
Andreas Wagner