[lustre-devel] [PATCH 12/15] lustre: osc: Don't get time for each page

[James Simmons]

From: Patrick Farrell <farr0186 at gmail.com>

Getting the time when each batch of pages starts is
sufficiently accurate, and ktime_get() is several % of the
CPU time when doing AIO + DIO.

This relies on previous patches in this series.

Measuring this in milliseconds/gigabyte lets us measure the
improvement in absolute terms, rather than just relative
terms.

This patch reduces i/o time in ms/GiB by:
Write: 17 ms/GiB
Read: 6 ms/GiB

Totals:
Write: 237 ms/GiB
Read: 223 ms/GiB

IOR:
mpirun -np 1  $IOR -w -r -t 64M -b 64G -o ./iorfile --posix.odirect
Without the patch:
write     4030 MiB/s
read      4468  MiB/s

With patch:
write     4326 MiB/s
read      4587 MiB/s

WC-bug-id: https://jira.whamcloud.com/browse/LU-13799
Lustre-commit: 485976ab451dd6708 ("LU-13799 osc: Don't get time for each page")
Signed-off-by: Patrick Farrell <farr0186 at gmail.com>
Reviewed-on: https://review.whamcloud.com/39437
Reviewed-by: Wang Shilong <wshilong at whamcloud.com>
Reviewed-by: Andreas Dilger <adilger at whamcloud.com>
Signed-off-by: James Simmons <jsimmons at infradead.org>
---
 fs/lustre/include/lustre_osc.h | 2 +-
 fs/lustre/osc/osc_io.c         | 3 ++-
 fs/lustre/osc/osc_page.c       | 4 ++--
 3 files changed, 5 insertions(+), 4 deletions(-)

diff --git a/fs/lustre/include/lustre_osc.h b/fs/lustre/include/lustre_osc.h
index 884ea59..208bb59 100644
--- a/fs/lustre/include/lustre_osc.h
+++ b/fs/lustre/include/lustre_osc.h
@@ -584,7 +584,7 @@ void osc_index2policy(union ldlm_policy_data *policy,
 		      pgoff_t start, pgoff_t end);
 void osc_lru_add_batch(struct client_obd *cli, struct list_head *list);
 void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
-		     enum cl_req_type crt, int brw_flags);
+		     enum cl_req_type crt, int brw_flags, ktime_t submit_time);
 int lru_queue_work(const struct lu_env *env, void *data);
 long osc_lru_shrink(const struct lu_env *env, struct client_obd *cli,
 		    long target, bool force);
diff --git a/fs/lustre/osc/osc_io.c b/fs/lustre/osc/osc_io.c
index 67fe85b..bd92b5d 100644
--- a/fs/lustre/osc/osc_io.c
+++ b/fs/lustre/osc/osc_io.c
@@ -132,6 +132,7 @@ int osc_io_submit(const struct lu_env *env, const struct cl_io_slice *ios,
 	unsigned int max_pages;
 	unsigned int ppc_bits; /* pages per chunk bits */
 	unsigned int ppc;
+	ktime_t submit_time = ktime_get();
 	bool sync_queue = false;
 
 	LASSERT(qin->pl_nr > 0);
@@ -195,7 +196,7 @@ int osc_io_submit(const struct lu_env *env, const struct cl_io_slice *ios,
 		oap->oap_async_flags |= ASYNC_COUNT_STABLE;
 		spin_unlock(&oap->oap_lock);
 
-		osc_page_submit(env, opg, crt, brw_flags);
+		osc_page_submit(env, opg, crt, brw_flags, submit_time);
 		list_add_tail(&oap->oap_pending_item, &list);
 
 		if (page->cp_sync_io)
diff --git a/fs/lustre/osc/osc_page.c b/fs/lustre/osc/osc_page.c
index 94db9d2..0f088fe 100644
--- a/fs/lustre/osc/osc_page.c
+++ b/fs/lustre/osc/osc_page.c
@@ -295,7 +295,7 @@ int osc_page_init(const struct lu_env *env, struct cl_object *obj,
  * transfer (i.e., transferred synchronously).
  */
 void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
-		     enum cl_req_type crt, int brw_flags)
+		     enum cl_req_type crt, int brw_flags, ktime_t submit_time)
 {
 	struct osc_io *oio = osc_env_io(env);
 	struct osc_async_page *oap = &opg->ops_oap;
@@ -316,7 +316,7 @@ void osc_page_submit(const struct lu_env *env, struct osc_page *opg,
 		oap->oap_cmd |= OBD_BRW_NOQUOTA;
 	}
 
-	opg->ops_submit_time = ktime_get();
+	opg->ops_submit_time = submit_time;
 	osc_page_transfer_get(opg, "transfer\0imm");
 	osc_page_transfer_add(env, opg, crt);
 }
-- 
1.8.3.1