ivampiresp/Spdk
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

nvme/perf: make the drain io in round robin.

Browse Source 
This patch is used to make the io completion
of different ns_ctx owned by each worker in
a round robin way.

Purpose: To avoid the timeout if there are many
ns_ctxes. For example, if each ns_ctxt connects
to remote NVMe-oF target with qpair. If there are
many qpairs, we may face the time out of some qpairs.

Change-Id: I477cb7436dc46ea498a26f990ed79001fa1bf2d6
Signed-off-by: Ziye Yang <optimistyzy@gmail.com>
Reviewed-on: https://review.gerrithub.io/c/438150
Tested-by: SPDK CI Jenkins <sys_sgci@intel.com>
Reviewed-by: Shuhei Matsumoto <shuhei.matsumoto.xt@hitachi.com>
Reviewed-by: Jim Harris <james.r.harris@intel.com>
This commit is contained in:
Ziye Yang 2018-12-24 13:20:07 +08:00 committed by Jim Harris
parent ea002f5068
commit 012a15837a
1 changed files with 22 additions and 15 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

37
examples/nvme/perf/perf.c
View File

@ -787,15 +787,6 @@ submit_io(struct ns_worker_ctx *ns_ctx, int queue_depth)
}
}
static void
drain_io(struct ns_worker_ctx *ns_ctx)
{
ns_ctx->is_draining = true;
while (ns_ctx->current_queue_depth > 0) {
check_io(ns_ctx);
}
}
static int
init_ns_worker_ctx(struct ns_worker_ctx *ns_ctx)
{
@ -854,6 +845,7 @@ work_fn(void *arg)
uint64_t tsc_end;
struct worker_thread *worker = (struct worker_thread *)arg;
struct ns_worker_ctx *ns_ctx = NULL;
uint32_t unfinished_ns_ctx;
printf("Starting thread on core %u\n", worker->lcore);
@ -893,12 +885,27 @@ work_fn(void *arg)
}
}
ns_ctx = worker->ns_ctx;
while (ns_ctx != NULL) {
drain_io(ns_ctx);
cleanup_ns_worker_ctx(ns_ctx);
ns_ctx = ns_ctx->next;
}
/* drain the io of each ns_ctx in round robin to make the fairness */
do {
unfinished_ns_ctx = 0;
ns_ctx = worker->ns_ctx;
while (ns_ctx != NULL) {
/* first time will enter into this if case */
if (!ns_ctx->is_draining) {
ns_ctx->is_draining = true;
}
if (ns_ctx->current_queue_depth > 0) {
check_io(ns_ctx);
if (ns_ctx->current_queue_depth == 0) {
cleanup_ns_worker_ctx(ns_ctx);
} else {
unfinished_ns_ctx++;
}
}
ns_ctx = ns_ctx->next;
}
} while (unfinished_ns_ctx > 0);
return 0;
}