On Mon, 18 Jan 2010, Linus Torvalds wrote:
>
> In particular:
>
> strace -s8 -T -o trace .libs/lt-blkid /dev/sdc1
> grep -v '<0.0' trace
>
> gives
>
> read(3, "\353<\220){'`j"..., 1024) = 1024 <1.024049>
> read(3, "\0\0\0\0\0\0\0\0"..., 16384) = 16384 <2.048790>
> close(3) = 0 <4.102118>
>
> and I have _no_ idea why that close() takes so long, but I'm guessing that
> it is flushing the page cache and has to wait for any read-ahead to
> finish.
>
> That is arguably a kernel problem, and I'll take a look at it. Although
> disabling read-ahead in blkid is probably a good thing regardless.
Here's a patch.
With this, I get:
read(3, "\353<\220){'`j"..., 1024) = 1024 <0.256857>
read(3, "\0\0\0\0\0\0\0\0"..., 16384) = 16384 <1.033671>
ie the reads themselves sped up (because we only read what we really need,
no read-ahead), and the overhead of "close()" obviously went away too.
So how I get
[root@EeePC misc-utils]# time ./blkid /dev/sdc1
/dev/sdc1: SEC_TYPE="msdos" TYPE="vfat"
real 0m1.342s
user 0m0.015s
sys 0m0.033s
which is a _lot_ better, and quite acceptable. It's not the 8kB read that
you claimed, but it's fine.
This improves the full-disk case too, but not enough:
[root@EeePC misc-utils]# time ./blkid /dev/sdc
real 0m4.968s
user 0m0.023s
sys 0m0.047s
I think it's still unacceptable to take five seconds to notice that it's
just a regular DOS partition thing, but it is certainly a lot faster than
it used to be.
Linus
---
From: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
Subject: Disable read-ahead when probing device files in blkid
Read-ahead doesn't work very well on device probing, and can hurt a lot
when we do essentially random accesses on very slow devices. So disable it
if possible.
Signed-off-by: Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx>
---
shlibs/blkid/src/probe.c | 4 ++++
1 files changed, 4 insertions(+), 0 deletions(-)
diff --git a/shlibs/blkid/src/probe.c b/shlibs/blkid/src/probe.c
index 2d3cafe..21791c1 100644
--- a/shlibs/blkid/src/probe.c
+++ b/shlibs/blkid/src/probe.c
@@ -640,6 +640,10 @@ int blkid_probe_set_device(blkid_probe pr, int fd,
pr->mode = 0;
pr->blkssz = 0;
+#ifdef POSIX_FADV_RANDOM
+ posix_fadvise(fd, 0, 0, POSIX_FADV_RANDOM);
+#endif
+
if (size)
pr->size = size;
else {
--
To unsubscribe from this list: send the line "unsubscribe util-linux-ng" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
