[PATCH -mm] readahead: partial sendfile fix

[PATCH -mm] readahead: partial sendfile fix

[Fengguang Wu]

Enable readahead to handle partially done read requests, e.g.

        sendfile(188, 1921, [1478592], 19553028) = 37440
        sendfile(188, 1921, [1516032], 19515588) = 28800
        sendfile(188, 1921, [1544832], 19486788) = 37440
        sendfile(188, 1921, [1582272], 19449348) = 14400
        sendfile(188, 1921, [1596672], 19434948) = 37440
        sendfile(188, 1921, [1634112], 19397508) = 37440

In the above strace log,
        - some lighttpd is doing _sequential_ reading
        - every sendfile() returns with only _partial_ work done

page_cache_readahead() expects that if it returns @next_index, it will be
called exactly at @next_index next time. That's not true here. So the pattern
will be falsely recognized as a random read trace.

Also documented in "Linux AIO Performance and Robustness for Enterprise
Workloads" section 3.5:

	  sendfile(fd, 0, 2GB, fd2) = 8192,
	    tells readahead about up to 128KB of the read
	  sendfile(fd, 8192, 2GB - 8192, fd2) = 8192,
	    tells readahead about 8KB - 132KB of the read
	  sendfile(fd, 16384, 2GB - 16384, fd2) = 8192,
	    tells readahead about 16KB-140KB of the read
	       ...
	This confuses the readahead logic about the I/O pattern which appears
	to be 0-128K, 8K-132K, 16K-140K instead of clear sequentiality from
	0-2GB that is really appropriate.

Retry based AIO shares the same read pattern and readahead problem.
In this case, simply disabling readahead on restarted aio is not a good option:
we still need to call into readahead in the rare case of (req_size > ra_max).

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
---
 mm/filemap.c   |    3 ---
 mm/readahead.c |    9 +++++++++
 2 files changed, 9 insertions(+), 3 deletions(-)

--- linux-2.6.20-rc6-mm3.orig/mm/readahead.c
+++ linux-2.6.20-rc6-mm3/mm/readahead.c
@@ -581,6 +581,15 @@ page_cache_readahead(struct address_spac
 	int sequential;
 
 	/*
+	 * A previous read request is partially completed,
+	 * causing the retried/continued read calls into us prematurely.
+	 */
+	if (ra->start < offset &&
+			offset < ra->prev_page &&
+				 ra->prev_page < ra->ahead_start + ra->ahead_size)
+		goto out;
+
+	/*
 	 * We avoid doing extra work and bogusly perturbing the readahead
 	 * window expansion logic.
 	 */
--- linux-2.6.20-rc6-mm3.orig/mm/filemap.c
+++ linux-2.6.20-rc6-mm3/mm/filemap.c
@@ -915,9 +915,6 @@ void do_generic_mapping_read(struct addr
 	if (!isize)
 		goto out;
 
-	if (unlikely(aio_restarted()))
-		next_index = last_index; /* Avoid repeat readahead */
-
 	end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
 	for (;;) {
 		struct page *page;

[PATCH -mm] readahead: partial sendfile fix

[Fengguang Wu]

Enable readahead to handle partially done read requests, e.g.

        sendfile(188, 1921, [1478592], 19553028) = 37440
        sendfile(188, 1921, [1516032], 19515588) = 28800
        sendfile(188, 1921, [1544832], 19486788) = 37440
        sendfile(188, 1921, [1582272], 19449348) = 14400
        sendfile(188, 1921, [1596672], 19434948) = 37440
        sendfile(188, 1921, [1634112], 19397508) = 37440

In the above strace log,
        - some lighttpd is doing _sequential_ reading
        - every sendfile() returns with only _partial_ work done

page_cache_readahead() expects that if it returns @next_index, it will be
called exactly at @next_index next time. That's not true here. So the pattern
will be falsely recognized as a random read trace.

Also documented in "Linux AIO Performance and Robustness for Enterprise
Workloads" section 3.5:

	  sendfile(fd, 0, 2GB, fd2) = 8192,
	    tells readahead about up to 128KB of the read
	  sendfile(fd, 8192, 2GB - 8192, fd2) = 8192,
	    tells readahead about 8KB - 132KB of the read
	  sendfile(fd, 16384, 2GB - 16384, fd2) = 8192,
	    tells readahead about 16KB-140KB of the read
	       ...
	This confuses the readahead logic about the I/O pattern which appears
	to be 0-128K, 8K-132K, 16K-140K instead of clear sequentiality from
	0-2GB that is really appropriate.

Retry based AIO shares the same read pattern and readahead problem.
In this case, simply disabling readahead on restarted aio is not a good option:
we still need to call into readahead in the rare case of (req_size > ra_max).

Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
---
 mm/filemap.c   |    3 ---
 mm/readahead.c |    9 +++++++++
 2 files changed, 9 insertions(+), 3 deletions(-)

--- linux-2.6.20-rc6-mm3.orig/mm/readahead.c
+++ linux-2.6.20-rc6-mm3/mm/readahead.c
@@ -581,6 +581,15 @@ page_cache_readahead(struct address_spac
 	int sequential;
 
 	/*
+	 * A previous read request is partially completed,
+	 * causing the retried/continued read calls into us prematurely.
+	 */
+	if (ra->start < offset &&
+			offset < ra->prev_page &&
+				 ra->prev_page < ra->ahead_start + ra->ahead_size)
+		goto out;
+
+	/*
 	 * We avoid doing extra work and bogusly perturbing the readahead
 	 * window expansion logic.
 	 */
--- linux-2.6.20-rc6-mm3.orig/mm/filemap.c
+++ linux-2.6.20-rc6-mm3/mm/filemap.c
@@ -915,9 +915,6 @@ void do_generic_mapping_read(struct addr
 	if (!isize)
 		goto out;
 
-	if (unlikely(aio_restarted()))
-		next_index = last_index; /* Avoid repeat readahead */
-
 	end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
 	for (;;) {
 		struct page *page;

Re: [PATCH -mm] readahead: partial sendfile fix

[Ram Pai]

On Sat, 2007-02-10 at 09:40 +0800, Fengguang Wu wrote:
> Enable readahead to handle partially done read requests, e.g.
> 
>         sendfile(188, 1921, [1478592], 19553028) = 37440
>         sendfile(188, 1921, [1516032], 19515588) = 28800
>         sendfile(188, 1921, [1544832], 19486788) = 37440
>         sendfile(188, 1921, [1582272], 19449348) = 14400
>         sendfile(188, 1921, [1596672], 19434948) = 37440
>         sendfile(188, 1921, [1634112], 19397508) = 37440
> 
> In the above strace log,
>         - some lighttpd is doing _sequential_ reading
>         - every sendfile() returns with only _partial_ work done
> 
> page_cache_readahead() expects that if it returns @next_index, it will
> be
> called exactly at @next_index next time. That's not true here. So the
> pattern
> will be falsely recognized as a random read trace.
> 
> Also documented in "Linux AIO Performance and Robustness for
> Enterprise
> Workloads" section 3.5:
> 
>           sendfile(fd, 0, 2GB, fd2) = 8192,
>             tells readahead about up to 128KB of the read
>           sendfile(fd, 8192, 2GB - 8192, fd2) = 8192,
>             tells readahead about 8KB - 132KB of the read
>           sendfile(fd, 16384, 2GB - 16384, fd2) = 8192,
>             tells readahead about 16KB-140KB of the read
>                ...
>         This confuses the readahead logic about the I/O pattern which
> appears
>         to be 0-128K, 8K-132K, 16K-140K instead of clear sequentiality
> from
>         0-2GB that is really appropriate.
> 
> Retry based AIO shares the same read pattern and readahead problem.
> In this case, simply disabling readahead on restarted aio is not a
> good option:
> we still need to call into readahead in the rare case of (req_size >
> ra_max).

The solution you proposed seems kludgy to me. If you determine that the
its a restarted aio, then start reading from where readahead had left
reading from earlier. To me a simple fix is:

-       if (unlikely(aio_restarted()))
-               next_index = last_index; /* Avoid repeat readahead */

+       if (unlikely(aio_restarted()))
+  		next_index = min(prev_index+1, last_index);


No? 
RP



> 
> Signed-off-by: Fengguang Wu <wfg@mail.ustc.edu.cn>
> ---
>  mm/filemap.c   |    3 ---
>  mm/readahead.c |    9 +++++++++
>  2 files changed, 9 insertions(+), 3 deletions(-)
> 
> --- linux-2.6.20-rc6-mm3.orig/mm/readahead.c
> +++ linux-2.6.20-rc6-mm3/mm/readahead.c
> @@ -581,6 +581,15 @@ page_cache_readahead(struct address_spac
>         int sequential;
> 
>         /*
> +        * A previous read request is partially completed,
> +        * causing the retried/continued read calls into us
> prematurely.
> +        */
> +       if (ra->start < offset &&
> +                       offset < ra->prev_page &&
> +                                ra->prev_page < ra->ahead_start +
> ra->ahead_size)
> +               goto out;
> +
> +       /*
>          * We avoid doing extra work and bogusly perturbing the
> readahead
>          * window expansion logic.
>          */
> --- linux-2.6.20-rc6-mm3.orig/mm/filemap.c
> +++ linux-2.6.20-rc6-mm3/mm/filemap.c
> @@ -915,9 +915,6 @@ void do_generic_mapping_read(struct addr
>         if (!isize)
>                 goto out;
> 
> -       if (unlikely(aio_restarted()))
> -               next_index = last_index; /* Avoid repeat readahead */




> -
>         end_index = (isize - 1) >> PAGE_CACHE_SHIFT;
>         for (;;) {
>                 struct page *page;
> 

Re: [PATCH -mm] readahead: partial sendfile fix

[Fengguang Wu]

Ram Pai,

Sorry for the long delay, I was just back from the winter vacation.

On Mon, Feb 12, 2007 at 11:49:10AM -0800, Ram Pai wrote:
> The solution you proposed seems kludgy to me. If you determine that the

I dislike it, either.

> its a restarted aio, then start reading from where readahead had left
> reading from earlier. To me a simple fix is:
> 
> -       if (unlikely(aio_restarted()))
> -               next_index = last_index; /* Avoid repeat readahead */
> 
> +       if (unlikely(aio_restarted()))
> +  		next_index = min(prev_index+1, last_index);
> 
> 
> No? 

Can be even simpler, if we _only_ want to fix the aio case:

+       if (unlikely(aio_restarted()))
+               next_index = prev_index + 1;

Regards,
Wu

Re: [PATCH -mm] readahead: partial sendfile fix

[Fengguang Wu]

Ram Pai,

Sorry for the long delay, I was just back from the winter vacation.

On Mon, Feb 12, 2007 at 11:49:10AM -0800, Ram Pai wrote:
> The solution you proposed seems kludgy to me. If you determine that the

I dislike it, either.

> its a restarted aio, then start reading from where readahead had left
> reading from earlier. To me a simple fix is:
> 
> -       if (unlikely(aio_restarted()))
> -               next_index = last_index; /* Avoid repeat readahead */
> 
> +       if (unlikely(aio_restarted()))
> +  		next_index = min(prev_index+1, last_index);
> 
> 
> No? 

Can be even simpler, if we _only_ want to fix the aio case:

+       if (unlikely(aio_restarted()))
+               next_index = prev_index + 1;

Regards,
Wu