readahead: introduce FMODE_RANDOM for POSIX_FADV_RANDOM
authorWu Fengguang <fengguang.wu@intel.com>
Fri, 5 Mar 2010 21:42:03 +0000 (13:42 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Sat, 6 Mar 2010 19:26:25 +0000 (11:26 -0800)
This fixes inefficient page-by-page reads on POSIX_FADV_RANDOM.

POSIX_FADV_RANDOM used to set ra_pages=0, which leads to poor performance:
a 16K read will be carried out in 4 _sync_ 1-page reads.

In other places, ra_pages==0 means
- it's ramfs/tmpfs/hugetlbfs/sysfs/configfs
- some IO error happened
where multi-page read IO won't help or should be avoided.

POSIX_FADV_RANDOM actually want a different semantics: to disable the
*heuristic* readahead algorithm, and to use a dumb one which faithfully
submit read IO for whatever application requests.

So introduce a flag FMODE_RANDOM for POSIX_FADV_RANDOM.

Note that the random hint is not likely to help random reads performance
noticeably.  And it may be too permissive on huge request size (its IO
size is not limited by read_ahead_kb).

In Quentin's report (http://lkml.org/lkml/2009/12/24/145), the overall
(NFS read) performance of the application increased by 313%!

Tested-by: Quentin Barnes <qbarnes+nfs@yahoo-inc.com>
Signed-off-by: Wu Fengguang <fengguang.wu@intel.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Andi Kleen <andi@firstfloor.org>
Cc: Steven Whitehouse <swhiteho@redhat.com>
Cc: David Howells <dhowells@redhat.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Trond Myklebust <Trond.Myklebust@netapp.com>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: <stable@kernel.org> [2.6.33.x]
Cc: <qbarnes+nfs@yahoo-inc.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/linux/fs.h
mm/fadvise.c
mm/readahead.c

index 45689621a85160af0d367eae8689737cc4629a7f..be87edcaba06be6c368c056524601ea26361f4bf 100644 (file)
@@ -87,6 +87,9 @@ struct inodes_stat_t {
  */
 #define FMODE_NOCMTIME         ((__force fmode_t)2048)
 
+/* Expect random access pattern */
+#define FMODE_RANDOM           ((__force fmode_t)4096)
+
 /*
  * The below are the various read and write types that we support. Some of
  * them include behavioral modifiers that send information down to the
index e43359214f6ff15020b6f05cc07aa5e6ddcb4162..8d723c9e8b75b316041ea564f5e243b79a94e2b7 100644 (file)
@@ -77,12 +77,20 @@ SYSCALL_DEFINE(fadvise64_64)(int fd, loff_t offset, loff_t len, int advice)
        switch (advice) {
        case POSIX_FADV_NORMAL:
                file->f_ra.ra_pages = bdi->ra_pages;
+               spin_lock(&file->f_lock);
+               file->f_mode &= ~FMODE_RANDOM;
+               spin_unlock(&file->f_lock);
                break;
        case POSIX_FADV_RANDOM:
-               file->f_ra.ra_pages = 0;
+               spin_lock(&file->f_lock);
+               file->f_mode |= FMODE_RANDOM;
+               spin_unlock(&file->f_lock);
                break;
        case POSIX_FADV_SEQUENTIAL:
                file->f_ra.ra_pages = bdi->ra_pages * 2;
+               spin_lock(&file->f_lock);
+               file->f_mode &= ~FMODE_RANDOM;
+               spin_unlock(&file->f_lock);
                break;
        case POSIX_FADV_WILLNEED:
                if (!mapping->a_ops->readpage) {
index 033bc135a41f3885cae863ca0c9f5185d6c5d665..337b20e946f6891ed9ae322d22c8e24a1b27f40e 100644 (file)
@@ -501,6 +501,12 @@ void page_cache_sync_readahead(struct address_space *mapping,
        if (!ra->ra_pages)
                return;
 
+       /* be dumb */
+       if (filp->f_mode & FMODE_RANDOM) {
+               force_page_cache_readahead(mapping, filp, offset, req_size);
+               return;
+       }
+
        /* do read-ahead */
        ondemand_readahead(mapping, ra, filp, false, offset, req_size);
 }