This helps performance on moderately dense random reads on SSD.
Transaction-Per-Second numbers provided by Taobao:
QPS case
-------------------------------------------------------
7536 disable context readahead totally
w/ patch: 7129 slower size rampup and start RA on the 3rd read
6717 slower size rampup
w/o patch: 5581 unmodified context readahead
Before, readahead will be started whenever reading page N+1 when it
happen to read N recently. After patch, we'll only start readahead
when *three* random reads happen to access pages N, N+1, N+2. The
probability of this happening is extremely low for pure random reads,
unless they are very dense, which actually deserves some readahead.
Also start with a smaller readahead window. The impact to interleaved
sequential reads should be small, because for a long run stream, the
the small readahead window rampup phase is negletable.
The context readahead actually benefits clustered random reads on HDD
whose seek cost is pretty high. However as SSD is increasingly used
for random read workloads it's better for the context readahead to
concentrate on interleaved sequential reads.
Another SSD rand read test from Miao
# file size: 2GB
# read IO amount: 625MB
sysbench --test=fileio \
--max-requests=10000 \
--num-threads=1 \
--file-num=1 \
--file-block-size=64K \
--file-test-mode=rndrd \
--file-fsync-freq=0 \
--file-fsync-end=off run
shows the performance of btrfs grows up from 69MB/s to 121MB/s,
ext4 from 104MB/s to 121MB/s.
Tested-by: Tao Ma <[email protected]>
Tested-by: Miao Xie <[email protected]>
Signed-off-by: Wu Fengguang <[email protected]>
---
mm/readahead.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
--- linux-next.orig/mm/readahead.c 2013-08-08 16:21:29.675286154 +0800
+++ linux-next/mm/readahead.c 2013-08-08 16:21:33.851286019 +0800
@@ -371,10 +371,10 @@ static int try_context_readahead(struct
size = count_history_pages(mapping, ra, offset, max);
/*
- * no history pages:
+ * not enough history pages:
* it could be a random read
*/
- if (!size)
+ if (size <= req_size)
return 0;
/*
@@ -385,8 +385,8 @@ static int try_context_readahead(struct
size *= 2;
ra->start = offset;
- ra->size = get_init_ra_size(size + req_size, max);
- ra->async_size = ra->size;
+ ra->size = min(size + req_size, max);
+ ra->async_size = 1;
return 1;
}
Hi, everyone
On Thu, 8 Aug 2013 16:54:18 +0800, Fengguang Wu wrote:
> This helps performance on moderately dense random reads on SSD.
>
> Transaction-Per-Second numbers provided by Taobao:
>
> QPS case
> -------------------------------------------------------
> 7536 disable context readahead totally
> w/ patch: 7129 slower size rampup and start RA on the 3rd read
> 6717 slower size rampup
> w/o patch: 5581 unmodified context readahead
>
> Before, readahead will be started whenever reading page N+1 when it
> happen to read N recently. After patch, we'll only start readahead
> when *three* random reads happen to access pages N, N+1, N+2. The
> probability of this happening is extremely low for pure random reads,
> unless they are very dense, which actually deserves some readahead.
>
> Also start with a smaller readahead window. The impact to interleaved
> sequential reads should be small, because for a long run stream, the
> the small readahead window rampup phase is negletable.
>
> The context readahead actually benefits clustered random reads on HDD
> whose seek cost is pretty high. However as SSD is increasingly used
> for random read workloads it's better for the context readahead to
> concentrate on interleaved sequential reads.
>
> Another SSD rand read test from Miao
>
> # file size: 2GB
> # read IO amount: 625MB
> sysbench --test=fileio \
> --max-requests=10000 \
> --num-threads=1 \
> --file-num=1 \
> --file-block-size=64K \
> --file-test-mode=rndrd \
> --file-fsync-freq=0 \
> --file-fsync-end=off run
>
> shows the performance of btrfs grows up from 69MB/s to 121MB/s,
> ext4 from 104MB/s to 121MB/s.
I did the same test on the hard disk recently,
for btrfs, there is ~5% regression(10.65MB/s -> 10.09MB/s),
for ext4, the performance grows up a bit.(9.98MB/s -> 10.04MB/s).
(I run the test for 4 times, and the above result is the average of the test.)
Any comment?
Thanks
Miao
>
> Tested-by: Tao Ma <[email protected]>
> Tested-by: Miao Xie <[email protected]>
> Signed-off-by: Wu Fengguang <[email protected]>
> ---
> mm/readahead.c | 8 ++++----
> 1 file changed, 4 insertions(+), 4 deletions(-)
>
> --- linux-next.orig/mm/readahead.c 2013-08-08 16:21:29.675286154 +0800
> +++ linux-next/mm/readahead.c 2013-08-08 16:21:33.851286019 +0800
> @@ -371,10 +371,10 @@ static int try_context_readahead(struct
> size = count_history_pages(mapping, ra, offset, max);
>
> /*
> - * no history pages:
> + * not enough history pages:
> * it could be a random read
> */
> - if (!size)
> + if (size <= req_size)
> return 0;
>
> /*
> @@ -385,8 +385,8 @@ static int try_context_readahead(struct
> size *= 2;
>
> ra->start = offset;
> - ra->size = get_init_ra_size(size + req_size, max);
> - ra->async_size = ra->size;
> + ra->size = min(size + req_size, max);
> + ra->async_size = 1;
>
> return 1;
> }
>
On Mon, Aug 19, 2013 at 09:59:09AM +0800, Miao Xie wrote:
> Hi, everyone
>
> On Thu, 8 Aug 2013 16:54:18 +0800, Fengguang Wu wrote:
> > This helps performance on moderately dense random reads on SSD.
> >
> > Transaction-Per-Second numbers provided by Taobao:
> >
> > QPS case
> > -------------------------------------------------------
> > 7536 disable context readahead totally
> > w/ patch: 7129 slower size rampup and start RA on the 3rd read
> > 6717 slower size rampup
> > w/o patch: 5581 unmodified context readahead
> >
> > Before, readahead will be started whenever reading page N+1 when it
> > happen to read N recently. After patch, we'll only start readahead
> > when *three* random reads happen to access pages N, N+1, N+2. The
> > probability of this happening is extremely low for pure random reads,
> > unless they are very dense, which actually deserves some readahead.
> >
> > Also start with a smaller readahead window. The impact to interleaved
> > sequential reads should be small, because for a long run stream, the
> > the small readahead window rampup phase is negletable.
> >
> > The context readahead actually benefits clustered random reads on HDD
> > whose seek cost is pretty high. However as SSD is increasingly used
> > for random read workloads it's better for the context readahead to
> > concentrate on interleaved sequential reads.
> >
> > Another SSD rand read test from Miao
> >
> > # file size: 2GB
> > # read IO amount: 625MB
> > sysbench --test=fileio \
> > --max-requests=10000 \
> > --num-threads=1 \
> > --file-num=1 \
> > --file-block-size=64K \
> > --file-test-mode=rndrd \
> > --file-fsync-freq=0 \
> > --file-fsync-end=off run
> >
> > shows the performance of btrfs grows up from 69MB/s to 121MB/s,
> > ext4 from 104MB/s to 121MB/s.
>
> I did the same test on the hard disk recently,
> for btrfs, there is ~5% regression(10.65MB/s -> 10.09MB/s),
> for ext4, the performance grows up a bit.(9.98MB/s -> 10.04MB/s).
> (I run the test for 4 times, and the above result is the average of the test.)
>
> Any comment?
Thanks for the tests! Minor regressions on the HDD cases are expected.
Since random read workloads are migrating to SSD as it becomes cheaper
and larger, it seems a good tradeoff to optimize for random read
performance on SSD.
Thanks,
Fengguang