Hi Jens,
I made changes to fio so we wld re-use the already populated io_u
buffer (when there is a non-random pattern) during writes. That way
only the header will be re-calculated for every I/O. This way the
buffer wld get populated in the beginning and as long as the
subsequent ios using the same io_u structure are writes and have same
or less block size, it wld get re-used. If any of the subsequent i/o
is a read or has a block size greater than the pre-filled one, then
the buffer is invalidated and will be re-filled at the next write.
Reason for this risky change: (Performance)
I tested this change on a tmpfs(with no swap backing), with the
following config file:
[sscan_write]
filename=/mytmpfs/datafile.tmp
rw=write
bs=64k
size=3G
ioengine=libaio
iodepth=1024
iodepth_low=512
runtime=10800
bwavgtime=5000
thread=1
do_verify=0
verify=meta
verify_pattern=0x55aaa55a
verify_interval=4k
continue_on_error=1
fio-1-41-6 gave 306MB/s and the new change had a performance of 1546MB/s
Side effects/Risks:
There is a risk with this fix, that if the buffer gets corrupted then
the subsequent writes will also be corrupt. I think for both
sequential writes and random writes (with verify, where the I/O log is
replayed) we shld be able to find the first I/O that started with the
corruption and if the buffer is getting corrupted, there are other
issues here.
Testing:
I have tested this fix with sequential write(verify)/random read write
mix combination(with verify).
I think I have taken care of most of the case, but please let me know
if there is anything I have missed. I have attached the patch along
with this email. I think the performance improvement outweighs the
risk associated with the fix. But I will let you decide if you wld
like to pick it up.
thanks
-radha
diff --git a/fio.c b/fio.c
index 2dab64e..896f797 100644
--- a/fio.c
+++ b/fio.c
@@ -831,6 +831,13 @@ static int init_io_u(struct thread_data *td)
if (td_write(td) && !td->o.refill_buffers)
io_u_fill_buffer(td, io_u, max_bs);
+ else if (td_write(td) && td->o.verify_pattern_bytes) {
+ /*
+ * Fill the buffer with the pattern if we are
+ * going to be doing writes.
+ */
+ fill_pattern(td, io_u->buf, max_bs, io_u);
+ }
}
io_u->index = i;
diff --git a/io_u.c b/io_u.c
index b2b7230..dc4473b 100644
--- a/io_u.c
+++ b/io_u.c
@@ -983,6 +983,14 @@ struct io_u *get_io_u(struct thread_data *td)
populate_verify_io_u(td, io_u);
else if (td->o.refill_buffers && io_u->ddir == DDIR_WRITE)
io_u_fill_buffer(td, io_u, io_u->xfer_buflen);
+ else if (io_u->ddir == DDIR_READ) {
+ /*
+ * Reset the buf_filled parameters so next time if the
+ * buffer is used for writes it is refilled.
+ */
+ io_u->buf_filled = 0;
+ io_u->buf_filled_len = 0;
+ }
}
/*
diff --git a/ioengine.h b/ioengine.h
index 91dd429..b599b61 100644
--- a/ioengine.h
+++ b/ioengine.h
@@ -43,6 +43,13 @@ struct io_u {
unsigned long long offset;
/*
+ * Parameters related to pre-filled buffers and
+ * their size to handle variable block sizes.
+ */
+ int buf_filled;
+ unsigned long buf_filled_len;
+
+ /*
* IO engine state, may be different from above when we get
* partial transfers / residual data counts
*/
diff --git a/verify.c b/verify.c
index 265bd55..73c1262 100644
--- a/verify.c
+++ b/verify.c
@@ -22,7 +22,7 @@
#include "crc/sha512.h"
#include "crc/sha1.h"
-static void fill_pattern(struct thread_data *td, void *p, unsigned int len)
+void fill_pattern(struct thread_data *td, void *p, unsigned int len, struct io_u *io_u)
{
switch (td->o.verify_pattern_bytes) {
case 0:
@@ -30,13 +30,24 @@ static void fill_pattern(struct thread_data *td, void *p, unsigned int len)
fill_random_buf(p, len);
break;
case 1:
+ if (io_u->buf_filled && io_u->buf_filled_len >= len) {
+ dprint(FD_VERIFY, "using already filled verify pattern b=0 len=%u\n", len);
+ return;
+ }
dprint(FD_VERIFY, "fill verify pattern b=0 len=%u\n", len);
memset(p, td->o.verify_pattern[0], len);
+ io_u->buf_filled = 1;
+ io_u->buf_filled_len = len;
break;
default: {
unsigned int i = 0, size = 0;
unsigned char *b = p;
+ if (io_u->buf_filled && io_u->buf_filled_len >= len) {
+ dprint(FD_VERIFY, "using already filled verify pattern b=%d len=%u\n",
+ td->o.verify_pattern_bytes, len);
+ return;
+ }
dprint(FD_VERIFY, "fill verify pattern b=%d len=%u\n",
td->o.verify_pattern_bytes, len);
@@ -47,6 +58,8 @@ static void fill_pattern(struct thread_data *td, void *p, unsigned int len)
memcpy(b+i, td->o.verify_pattern, size);
i += size;
}
+ io_u->buf_filled = 1;
+ io_u->buf_filled_len = len;
break;
}
}
@@ -675,7 +688,7 @@ void populate_verify_io_u(struct thread_data *td, struct io_u *io_u)
if (td->o.verify == VERIFY_NULL)
return;
- fill_pattern(td, p, io_u->buflen);
+ fill_pattern(td, p, io_u->buflen, io_u);
hdr_inc = io_u->buflen;
if (td->o.verify_interval)
diff --git a/verify.h b/verify.h
index a4a8cfe..29c4b46 100644
--- a/verify.h
+++ b/verify.h
@@ -69,6 +69,7 @@ extern void populate_verify_io_u(struct thread_data *, struct io_u *);
extern int __must_check get_next_verify(struct thread_data *td, struct io_u *);
extern int __must_check verify_io_u(struct thread_data *, struct io_u *);
extern int verify_io_u_async(struct thread_data *, struct io_u *);
+extern void fill_pattern(struct thread_data *td, void *p, unsigned int len, struct io_u *io_u);
/*
* Async verify offload
