The following changes since commit 9a0ced5a97ae9fde87dae1a3d762b1bc090e8786:
sync engine: guard pvsync engine register/unregister (2013-05-16 20:50:26 +0200)
are available in the git repository at:
git://git.kernel.dk/fio.git master
Jens Axboe (4):
Ensure that we have no IO pending when sleeping
Merge branch 'master' of ssh://git.kernel.dk/data/git/fio
Update HOWTO and man page for thinktime blocks impact on queue depth
flow: quiesce pending IO before sleep
HOWTO | 8 ++++++--
backend.c | 2 ++
fio.1 | 7 ++++++-
flow.c | 5 ++++-
io_u.c | 33 +++++++++++++++++++--------------
ioengine.h | 1 +
6 files changed, 38 insertions(+), 18 deletions(-)
---
Diff of recent changes:
diff --git a/HOWTO b/HOWTO
index 9b2edb3..794728f 100644
--- a/HOWTO
+++ b/HOWTO
@@ -830,11 +830,15 @@ thinktime_spin=int
to sleeping for the rest of the period specified by
thinktime.
-thinktime_blocks
+thinktime_blocks=int
Only valid if thinktime is set - control how many blocks
to issue, before waiting 'thinktime' usecs. If not set,
defaults to 1 which will make fio wait 'thinktime' usecs
- after every block.
+ after every block. This effectively makes any queue depth
+ setting redundant, since no more than 1 IO will be queued
+ before we have to complete it and do our thinktime. In
+ other words, this setting effectively caps the queue depth
+ if the latter is larger.
rate=int Cap the bandwidth used by this job. The number is in bytes/sec,
the normal suffix rules apply. You can use rate=500k to limit
diff --git a/backend.c b/backend.c
index 89ffee1..6b6da67 100644
--- a/backend.c
+++ b/backend.c
@@ -822,6 +822,8 @@ sync_done:
if (!(b % td->o.thinktime_blocks)) {
int left;
+ io_u_quiesce(td);
+
if (td->o.thinktime_spin)
usec_spin(td->o.thinktime_spin);
diff --git a/fio.1 b/fio.1
index f54ea34..91020b2 100644
--- a/fio.1
+++ b/fio.1
@@ -702,7 +702,12 @@ Pretend to spend CPU time for given number of microseconds, sleeping the rest
of the time specified by \fBthinktime\fR. Only valid if \fBthinktime\fR is set.
.TP
.BI thinktime_blocks \fR=\fPint
-Number of blocks to issue before waiting \fBthinktime\fR microseconds.
+Only valid if thinktime is set - control how many blocks to issue, before
+waiting \fBthinktime\fR microseconds. If not set, defaults to 1 which will
+make fio wait \fBthinktime\fR microseconds after every block. This
+effectively makes any queue depth setting redundant, since no more than 1 IO
+will be queued before we have to complete it and do our thinktime. In other
+words, this setting effectively caps the queue depth if the latter is larger.
Default: 1.
.TP
.BI rate \fR=\fPint
diff --git a/flow.c b/flow.c
index b7a2fb1..f9d868d 100644
--- a/flow.c
+++ b/flow.c
@@ -23,8 +23,11 @@ int flow_threshold_exceeded(struct thread_data *td)
sign = td->o.flow > 0 ? 1 : -1;
if (sign * flow->flow_counter > td->o.flow_watermark) {
- if (td->o.flow_sleep)
+ if (td->o.flow_sleep) {
+ io_u_quiesce(td);
usleep(td->o.flow_sleep);
+ }
+
return 1;
}
diff --git a/io_u.c b/io_u.c
index 2cf2b8d..f0b6170 100644
--- a/io_u.c
+++ b/io_u.c
@@ -511,6 +511,24 @@ static inline enum fio_ddir get_rand_ddir(struct thread_data *td)
return DDIR_WRITE;
}
+void io_u_quiesce(struct thread_data *td)
+{
+ /*
+ * We are going to sleep, ensure that we flush anything pending as
+ * not to skew our latency numbers.
+ *
+ * Changed to only monitor 'in flight' requests here instead of the
+ * td->cur_depth, b/c td->cur_depth does not accurately represent
+ * io's that have been actually submitted to an async engine,
+ * and cur_depth is meaningless for sync engines.
+ */
+ while (td->io_u_in_flight) {
+ int fio_unused ret;
+
+ ret = io_u_queued_complete(td, 1, NULL);
+ }
+}
+
static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir)
{
enum fio_ddir odir = ddir ^ 1;
@@ -547,20 +565,7 @@ static enum fio_ddir rate_ddir(struct thread_data *td, enum fio_ddir ddir)
} else
usec = td->rate_pending_usleep[ddir];
- /*
- * We are going to sleep, ensure that we flush anything pending as
- * not to skew our latency numbers.
- *
- * Changed to only monitor 'in flight' requests here instead of the
- * td->cur_depth, b/c td->cur_depth does not accurately represent
- * io's that have been actually submitted to an async engine,
- * and cur_depth is meaningless for sync engines.
- */
- while (td->io_u_in_flight) {
- int fio_unused ret;
-
- ret = io_u_queued_complete(td, 1, NULL);
- }
+ io_u_quiesce(td);
fio_gettime(&t, NULL);
usec_sleep(td, usec);
diff --git a/ioengine.h b/ioengine.h
index 362ab3e..0be905f 100644
--- a/ioengine.h
+++ b/ioengine.h
@@ -196,6 +196,7 @@ extern void requeue_io_u(struct thread_data *, struct io_u **);
extern int __must_check io_u_sync_complete(struct thread_data *, struct io_u *, uint64_t *);
extern int __must_check io_u_queued_complete(struct thread_data *, int, uint64_t *);
extern void io_u_queued(struct thread_data *, struct io_u *);
+extern void io_u_quiesce(struct thread_data *);
extern void io_u_log_error(struct thread_data *, struct io_u *);
extern void io_u_mark_depth(struct thread_data *, unsigned int);
extern void fill_io_buffer(struct thread_data *, void *, unsigned int, unsigned int);
--
To unsubscribe from this list: send the line "unsubscribe fio" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
