lkml.org 
[lkml]   [2010]   [Sep]   [27]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRequest starvation with CFQ
  Hi,

when helping Lennart with answering some questions, I've spotted the
following problem (at least I think it's a problem ;): The thing is that
CFQ schedules how requests should be dispatched but does not in any
significant way limit to whom requests get allocated. Given we have a
quite limited pool of available requests it can happen that processes
will be actually starved not waiting for disk but waiting for requests
getting allocated and any IO scheduling priorities or classes will not
have serious effect.
A pathological example I've tried below:
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <sys/stat.h>

int main(void)
{
int fd = open("/dev/vdb", O_RDONLY);
int loop = 0;

if (fd < 0) {
perror("open");
exit(1);
}
while (1) {
if (loop % 100 == 0)
printf("Loop %d\n", loop);
posix_fadvise(fd, (random() * 4096) % 1000204886016ULL, 4096, POSIX_FADV_WILLNEED);
loop++;
}
}

This program will just push as many requests as possible to the block
layer and does not wait for any IO. Thus it will basically ignore any
decisions about when requests get dispatched. BTW, don't get distracted
by the fact that the program operates directly on the device, that is just
for simplicity. Large enough file would work the same way.
Even though I run this program with ionice -c 3, I still see that any
other IO to the device is basically stalled. When I look at the block
traces, I indeed see that what happens is that the above program submits
requests until there are no more available:
...
254,16 2 802 1.411285520 2563 Q R 696733184 + 8 [random_read]
254,16 2 803 1.411314880 2563 G R 696733184 + 8 [random_read]
254,16 2 804 1.411338220 2563 I R 696733184 + 8 [random_read]
254,16 2 805 1.411415040 2563 Q R 1006864600 + 8 [random_read]
254,16 2 806 1.411441620 2563 S R 1006864600 + 8 [random_read]

during and after that IO happens:
254,16 3 31 1.417898030 0 C R 345134640 + 8 [0]
254,16 3 32 1.418171910 0 D R 1524771568 + 8 [swapper]
254,16 0 33 1.432317140 0 C R 1524771568 + 8 [0]
254,16 0 34 1.432597000 0 D R 1077270768 + 8 [swapper]
...
254,16 0 35 1.503238050 0 C R 33633744 + 8 [0]
254,16 0 36 1.503558290 0 D R 22178968 + 8 [swapper]

and the other program comes with IO and gets stalled:
254,16 1 39 1.508843180 2564 A RM 12346 + 8 <- (254,17) 12312
254,16 1 40 1.508876520 2564 Q RM 12346 + 8 [ls]
254,16 1 41 1.508905140 2564 S RM 12346 + 8 [ls]
...
IO is still running:
254,16 2 807 1.512081560 0 C R 22178968 + 8 [0]
254,16 2 808 1.512365010 0 D R 475025688 + 8 [swapper]
254,16 3 35 1.522113270 0 C R 475025688 + 8 [0]
254,16 3 36 1.522390779 0 D R 697010128 + 8 [swapper]
254,16 4 33 1.531443760 0 C R 697010128 + 8 [0]
...
random reader even gets to submitting more requests:
254,16 2 815 1.785734950 2563 G R 1006864600 + 8 [random_read]
254,16 2 816 1.785752290 2563 I R 1006864600 + 8 [random_read]
254,16 2 817 1.785825880 2563 Q R 832683552 + 8 [random_read]
254,16 2 818 1.785850890 2563 G R 832683552 + 8 [random_read]
254,16 2 819 1.785874610 2563 I R 832683552 + 8 [random_read]
...
and finally our program gets to adding it's request as well:
254,16 1 60 2.160884040 2564 G RM 12346 + 8 [ls]
254,16 1 61 2.160914700 2564 I R 12346 + 8 [ls]
254,16 1 62 2.161142170 2564 D R 12346 + 8 [ls]
254,16 1 63 2.161233670 2564 U N [ls] 128

I can provide the full traces for download if someone is interested
in some part I didn't include here. The kernel is 2.6.36-rc4.
Now I agree that the above program is about as bad as it can get but
Lennart would like to implement readahead during boot on background and
I believe that could starve other IO in a similar way. So any idea how
to solve this? To me it seems as if we also needed to somehow limit the
number of allocated requests per cfqq but OTOH we have to be really careful
to not harm common workloads where we benefit from having lots of requests
queued...

Honza
--
Jan Kara <jack@suse.cz>
SUSE Labs, CR


\
 
 \ /
  Last update: 2010-09-27 21:03    [W:0.068 / U:0.560 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site