Main changes:
- By default, allocate all hugepages on node0. On NUMA-aware systems,
processes will most often use default policy with local node as the
preffered one for allocations. This usually is node0, thus splitting
hugepages evenly across the nodes, in default setup, would force
allocations from a remote node, impacting overall performance (in
case hugepages on node0 run out). See 68740678e1 as a reference.
- Introduce HUGE_EVEN_ALLOC - force setup.sh to evenly distribute
hugepages across all the nodes.
- Introduce HUGEPGSZ - overrides default page size
- Introduce CLEAR_HUGE - remove all hugepages on demand before
allocation is performed.
- HUGENODE - this now can be a list of nodes to use. NRHUGE will be
set across all the nodes from the list.
Change-Id: I084829edde3c416e7fc6b7b6abe369cc0631fcd7
Signed-off-by: Michal Berger <michalx.berger@intel.com>
Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/5042
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Community-CI: Mellanox Build Bot
Reviewed-by: Jim Harris <james.r.harris@intel.com>
Reviewed-by: Tomasz Zawadzki <tomasz.zawadzki@intel.com>
2b80955c09