+5
−2
Loading
In current version we reap all avaialbe RDMA CM events, at the big scale it may cause imbalance in IO handling when we receive many events in 1 poll (e.g. RDMA_CM_EVENT_DISCONNECTED) and spend much time on hanlding these events. To avoid it, reap up to 32 events per acceptor poll. Signed-off-by:Alexey Marchuk <alexeymar@nvidia.com> Change-Id: I638a826762f164eecbd37c3cccf12246390b9bbb Reviewed-on: https://review.spdk.io/gerrit/c/spdk/spdk/+/22272 Reviewed-by:
John Kariuki <John.Kariuki@amd.com> Tested-by:
SPDK CI Jenkins <sys_sgci@intel.com> Reviewed-by:
Jim Harris <jim.harris@samsung.com> Reviewed-by:
Shuhei Matsumoto <smatsumoto@nvidia.com> Community-CI: Mellanox Build Bot