RFC/Patch: Make NFS Readahead tunable

RFC/Patch: Make NFS Readahead tunable

[Martin Knoblauch]

[-- Attachment #1: Type: text/plain, Size: 3954 bytes --]

Hi,

 while tracking down a very interesting interaction between Sun/SAM-FS and Linux NFS clients, we found out that the value of NFS_MAX_READ_AHEAD is to agressive/big for the specific use-case.

 For testing, instead of always recompiling the kernel with different values, I came up with the following patch. It introduces a tunable  "/proc/sys/fs/nfs/nfs_ra_factor" with possible values between 0-15.

 Not sure whether it is actually a good thing to have. Better would be to set the read-ahead factor per filesystem via a mount option.

 The patch is against 2.6.24. It applies with offsets against 2.6.25-rc9. In case my mail client messes up the whitespace, the patch is also attached.

Signed-off-by: Martin Knoblauch <knobi-Ys4E+72pFW0hFhg+JK9F0w@public.gmane.org>

diff -ru linux-2.6.24-orig/fs/nfs/client.c linux-2.6.24-nfs_ra/fs/nfs/client.c
--- linux-2.6.24-orig/fs/nfs/client.c   2008-01-24 23:58:37.000000000 +0100
+++ linux-2.6.24-nfs_ra/fs/nfs/client.c 2008-04-14 16:44:22.000000000 +0200
@@ -601,6 +601,11 @@
 }

 /*
+ * NFS Client Read-Ahead factor
+*/
+unsigned int nfs_ra_factor;
+
+/*
  * Load up the server record from information gained in an fsinfo record
  */
 static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo *fsinfo)
@@ -625,7 +630,9 @@
                server->rsize = NFS_MAX_FILE_IO_SIZE;
        server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;

-       server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
+       printk(KERN_INFO "nfs_server_set_fsinfo: rsize, wsize, rpages, nfs_ra_factor, ra_pages: %d %d %d %d %d\n",
+               server->rsize,server->wsize,server->rpages,nfs_ra_factor,server->rpages * nfs_ra_factor);
+       server->backing_dev_info.ra_pages = server->rpages * nfs_ra_factor;

        if (server->wsize > max_rpc_payload)
                server->wsize = max_rpc_payload;
@@ -1359,6 +1366,8 @@
 {
        struct proc_dir_entry *p;

+       nfs_ra_factor = NFS_MAX_READAHEAD;
+
        proc_fs_nfs = proc_mkdir("nfsfs", proc_root_fs);
        if (!proc_fs_nfs)
                goto error_0;
diff -ru linux-2.6.24-orig/fs/nfs/sysctl.c linux-2.6.24-nfs_ra/fs/nfs/sysctl.c
--- linux-2.6.24-orig/fs/nfs/sysctl.c   2008-01-24 23:58:37.000000000 +0100
+++ linux-2.6.24-nfs_ra/fs/nfs/sysctl.c 2008-04-14 16:15:03.000000000 +0200
@@ -14,9 +14,12 @@
 #include <linux/nfs_fs.h>

 #include "callback.h"
+#include "internal.h"

 static const int nfs_set_port_min = 0;
 static const int nfs_set_port_max = 65535;
+static const unsigned int min_nfs_ra_factor = 0;
+static const unsigned int max_nfs_ra_factor = NFS_MAX_READAHEAD;
 static struct ctl_table_header *nfs_callback_sysctl_table;

 static ctl_table nfs_cb_sysctls[] = {
@@ -58,6 +61,16 @@
                .mode           = 0644,
                .proc_handler   = &proc_dointvec,
        },
+       {
+               .ctl_name = CTL_UNNUMBERED,
+               .procname = "nfs_ra_factor",
+               .data = &nfs_ra_factor,
+               .maxlen = sizeof(unsigned int),
+               .mode = 0644,
+               .proc_handler = &proc_dointvec_minmax,
+               .extra1 = (unsigned int *)&min_nfs_ra_factor,
+               .extra2 = (unsigned int *)&max_nfs_ra_factor,
+       },
        { .ctl_name = 0 }
 };

diff -ru linux-2.6.24-orig/include/linux/nfs_fs.h linux-2.6.24-nfs_ra/include/linux/nfs_fs.h
--- linux-2.6.24-orig/include/linux/nfs_fs.h    2008-01-24 23:58:37.000000000 +0100
+++ linux-2.6.24-nfs_ra/include/linux/nfs_fs.h  2008-04-14 15:58:57.000000000 +0200
@@ -415,6 +415,11 @@
 extern void nfs_writedata_release(void *);

 /*
+ * linux/fs/nfs/client.c
+*/
+extern unsigned int nfs_ra_factor;
+
+/*
  * Try to write back everything synchronously (but check the
  * return value!)
  */


Cheers
 Martin
------------------------------------------------------
Martin Knoblauch
email: k n o b i AT knobisoft DOT de
www:   http://www.knobisoft.de



[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #2: nfs_ra-2.6.24.patch --]
[-- Type: text/x-patch; name="nfs_ra-2.6.24.patch", Size: 2751 bytes --]

diff -ru linux-2.6.24-orig/fs/nfs/client.c linux-2.6.24-nfs_ra/fs/nfs/client.c
--- linux-2.6.24-orig/fs/nfs/client.c	2008-01-24 23:58:37.000000000 +0100
+++ linux-2.6.24-nfs_ra/fs/nfs/client.c	2008-04-14 16:44:22.000000000 +0200
@@ -601,6 +601,11 @@
 }
 
 /*
+ * NFS Client Read-Ahead factor
+*/
+unsigned int nfs_ra_factor;
+
+/*
  * Load up the server record from information gained in an fsinfo record
  */
 static void nfs_server_set_fsinfo(struct nfs_server *server, struct nfs_fsinfo *fsinfo)
@@ -625,7 +630,9 @@
 		server->rsize = NFS_MAX_FILE_IO_SIZE;
 	server->rpages = (server->rsize + PAGE_CACHE_SIZE - 1) >> PAGE_CACHE_SHIFT;
 
-	server->backing_dev_info.ra_pages = server->rpages * NFS_MAX_READAHEAD;
+	printk(KERN_INFO "nfs_server_set_fsinfo: rsize, wsize, rpages, nfs_ra_factor, ra_pages: %d %d %d %d %d\n",
+		server->rsize,server->wsize,server->rpages,nfs_ra_factor,server->rpages * nfs_ra_factor);
+	server->backing_dev_info.ra_pages = server->rpages * nfs_ra_factor;
 
 	if (server->wsize > max_rpc_payload)
 		server->wsize = max_rpc_payload;
@@ -1359,6 +1366,8 @@
 {
 	struct proc_dir_entry *p;
 
+	nfs_ra_factor = NFS_MAX_READAHEAD;
+
 	proc_fs_nfs = proc_mkdir("nfsfs", proc_root_fs);
 	if (!proc_fs_nfs)
 		goto error_0;
diff -ru linux-2.6.24-orig/fs/nfs/sysctl.c linux-2.6.24-nfs_ra/fs/nfs/sysctl.c
--- linux-2.6.24-orig/fs/nfs/sysctl.c	2008-01-24 23:58:37.000000000 +0100
+++ linux-2.6.24-nfs_ra/fs/nfs/sysctl.c	2008-04-14 16:15:03.000000000 +0200
@@ -14,9 +14,12 @@
 #include <linux/nfs_fs.h>
 
 #include "callback.h"
+#include "internal.h"
 
 static const int nfs_set_port_min = 0;
 static const int nfs_set_port_max = 65535;
+static const unsigned int min_nfs_ra_factor = 0;
+static const unsigned int max_nfs_ra_factor = NFS_MAX_READAHEAD;
 static struct ctl_table_header *nfs_callback_sysctl_table;
 
 static ctl_table nfs_cb_sysctls[] = {
@@ -58,6 +61,16 @@
 		.mode		= 0644,
 		.proc_handler	= &proc_dointvec,
 	},
+	{
+		.ctl_name = CTL_UNNUMBERED,
+		.procname = "nfs_ra_factor",
+		.data = &nfs_ra_factor,
+		.maxlen = sizeof(unsigned int),
+		.mode = 0644,
+		.proc_handler = &proc_dointvec_minmax,
+		.extra1 = (unsigned int *)&min_nfs_ra_factor,
+		.extra2 = (unsigned int *)&max_nfs_ra_factor,
+	},
 	{ .ctl_name = 0 }
 };
 
diff -ru linux-2.6.24-orig/include/linux/nfs_fs.h linux-2.6.24-nfs_ra/include/linux/nfs_fs.h
--- linux-2.6.24-orig/include/linux/nfs_fs.h	2008-01-24 23:58:37.000000000 +0100
+++ linux-2.6.24-nfs_ra/include/linux/nfs_fs.h	2008-04-14 15:58:57.000000000 +0200
@@ -415,6 +415,11 @@
 extern void nfs_writedata_release(void *);
 
 /*
+ * linux/fs/nfs/client.c
+*/
+extern unsigned int nfs_ra_factor;
+
+/*
  * Try to write back everything synchronously (but check the
  * return value!)
  */

Re: RFC/Patch: Make NFS Readahead tunable

[Trond Myklebust]

On Tue, 2008-04-15 at 06:39 -0700, Martin Knoblauch wrote:
> Hi,
> 
>  while tracking down a very interesting interaction between Sun/SAM-FS and Linux NFS clients, we found out that the value of NFS_MAX_READ_AHEAD is to agressive/big for the specific use-case.
> 
>  For testing, instead of always recompiling the kernel with different values, I came up with the following patch. It introduces a tunable  "/proc/sys/fs/nfs/nfs_ra_factor" with possible values between 0-15.
> 
>  Not sure whether it is actually a good thing to have. Better would be to set the read-ahead factor per filesystem via a mount option.
> 
>  The patch is against 2.6.24. It applies with offsets against 2.6.25-rc9. In case my mail client messes up the whitespace, the patch is also attached.

NFS_MAX_READ_AHEAD just sets an upper limit on the standard vfs/mm-level
page readahead algorithm. If you need finer control over that readahead
algorithm, then the kernel already has full support for the
posix_fadvise() system call.

Cheers
  Trond

Re: RFC/Patch: Make NFS Readahead tunable

[Martin Knoblauch]

----- Original Message ----
> From: Trond Myklebust <trond.myklebust@fys.uio.no>
> To: Martin Knoblauch <spamtrap-Ys4E+72pFW0hFhg+JK9F0w@public.gmane.org>
> Cc: linux-nfs list <linux-nfs@vger.kernel.org>
> Sent: Tuesday, April 15, 2008 6:12:32 PM
> Subject: Re: RFC/Patch: Make NFS Readahead tunable
> 
> 
> On Tue, 2008-04-15 at 06:39 -0700, Martin Knoblauch wrote:
> > Hi,
> > 
> >  while tracking down a very interesting interaction between Sun/SAM-FS and 
> Linux NFS clients, we found out that the value of NFS_MAX_READ_AHEAD is to 
> agressive/big for the specific use-case.
> > 
> >  For testing, instead of always recompiling the kernel with different values, 
> I came up with the following patch. It introduces a tunable  
> "/proc/sys/fs/nfs/nfs_ra_factor" with possible values between 0-15.
> > 
> >  Not sure whether it is actually a good thing to have. Better would be to set 
> the read-ahead factor per filesystem via a mount option.
> > 
> >  The patch is against 2.6.24. It applies with offsets against 2.6.25-rc9. In 
> case my mail client messes up the whitespace, the patch is also attached.
> 
> NFS_MAX_READ_AHEAD just sets an upper limit on the standard vfs/mm-level
> page readahead algorithm. If you need finer control over that readahead
> algorithm, then the kernel already has full support for the
> posix_fadvise() system call.
>

 But to use  posix_fadvise I would need control over the application. Correct? This is not given in the case I am actually dealing with. The NFS-server is Solaris-10/Sparc/SAM-FS. It turns out that *any* access to an "offline" file from a Linux-NFS client messes (and I mean messes) up the performance of the stager process if the read-ahead window is larger that 2 or three time rsize. Now - I have a pretty clear opinion of "who is at fault", but doing a band-aid on the Linux side may just be the easier thing, at least short/mid-term. Not sure how important the end-customer really is to the makers of SAM-FS.

 In general, of course controlling the readahead behaviour from the application is the better solution. anyway, the patch is simple enough to carry around.

Martin

> Cheers
>   Trond
> 
> 
>