Currently the socklnd driver doesn't support CPT affinity for its peers. Binding peers to a specific CPT and memory allocated to the NUMA node belonging to the CPT should give a performance boost. Signed-off-by: James Simmons <uja.ornl@xxxxxxxxx> Intel-bug-id: https://jira.hpdd.intel.com/browse/LU-7245 Reviewed-on: http://review.whamcloud.com/16710 Reviewed-by: Olaf Weber <olaf@xxxxxxx> Reviewed-by: Amir Shehata <amir.shehata@xxxxxxxxx> Reviewed-by: Oleg Drokin <oleg.drokin@xxxxxxxxx> --- .../staging/lustre/lnet/klnds/socklnd/socklnd.c | 3 ++- 1 files changed, 2 insertions(+), 1 deletions(-) diff --git a/drivers/staging/lustre/lnet/klnds/socklnd/socklnd.c b/drivers/staging/lustre/lnet/klnds/socklnd/socklnd.c index d843082..2c83b95 100644 --- a/drivers/staging/lustre/lnet/klnds/socklnd/socklnd.c +++ b/drivers/staging/lustre/lnet/klnds/socklnd/socklnd.c @@ -102,6 +102,7 @@ ksocknal_destroy_route(ksock_route_t *route) static int ksocknal_create_peer(ksock_peer_t **peerp, lnet_ni_t *ni, lnet_process_id_t id) { + int cpt = lnet_cpt_of_nid(id.nid); ksock_net_t *net = ni->ni_data; ksock_peer_t *peer; @@ -109,7 +110,7 @@ ksocknal_create_peer(ksock_peer_t **peerp, lnet_ni_t *ni, lnet_process_id_t id) LASSERT(id.pid != LNET_PID_ANY); LASSERT(!in_interrupt()); - LIBCFS_ALLOC(peer, sizeof(*peer)); + LIBCFS_CPT_ALLOC(peer, lnet_cpt_table(), cpt, sizeof(*peer)); if (!peer) return -ENOMEM; -- 1.7.1 _______________________________________________ devel mailing list devel@xxxxxxxxxxxxxxxxxxxxxx http://driverdev.linuxdriverproject.org/mailman/listinfo/driverdev-devel
