[LTP] [PATCH] readahead02: estimate max readahead size

[Jan Stancek <jstancek@redhat.com>]

----- Original Message -----
> From: "Cyril Hrubis" <chrubis@suse.cz>
> To: "Jan Stancek" <jstancek@redhat.com>
> Cc: ltp@lists.linux.it
> Sent: Monday, 10 October, 2016 4:16:52 PM
> Subject: Re: [PATCH] readahead02: estimate max readahead size
> 
> Hi!
> > +		do {
> > +			TEST(readahead(fd, offset, fsize - offset));
> > +			if (TEST_RETURN != 0) {
> > +				check_ret(0);
> >  				break;
> > -		}
> > -		check_ret(0);
> > +			}
> > +
> > +			/* estimate max readahead size based on first call */
> > +			if (!max_ra_estimate) {
> > +				*cached = get_cached_size();
> > +				if (*cached > cached_start) {
> > +					max_ra_estimate = ((*cached - cached_start)
> > +						* 1024 & (~(pagesize - 1)));
> 
> I'm just curious, why do we round the value down to pagesize?

Not sure  why I left it in, my previous approach was checking increase
in each loop, but that had issues, because it varied a lot among iteration
(whole system contributes to that number).
Should be safe to drop.

> 
> 
> 
> I've tried this on the /tmp on Btrfs and got max_ra_estimage nearly ten times
> greater than the sysfs value. But apart from that the measured times are
> pretty
> much comparable with the previous version, which suggests that maximal
> readahead size on Btrfs is set to ~4M and the limit per device is not
> applicable.
> 
> Did you get the estimate close to the value of the sysfs file in your
> environment?

I do (on 4.8 x86 KVM guest):
# cat /sys/devices/virtual/bdi/253:1/read_ahead_kb
4096
readahead02    0  TINFO  :  max ra estimate: 4145152

Are you using upstream kernel or a patched one?

Regards,
Jan

> 
> Your version:
> 
> readahead02    0  TINFO  :  creating test file of size: 67108864
> readahead02    0  TINFO  :  read_testfile(0)
> readahead02    0  TINFO  :  read_testfile(1)
> readahead02    0  TINFO  :  max ra estimate: 4177920
> readahead02    0  TINFO  :  readahead calls made: 17
> readahead02    1  TPASS  :  offset is still at 0 as expected
> readahead02    0  TINFO  :  read_testfile(0) took: 103324 usec
> readahead02    0  TINFO  :  read_testfile(1) took: 22291 usec
> readahead02    0  TINFO  :  read_testfile(0) read: 67190784 bytes
> readahead02    0  TINFO  :  read_testfile(1) read: 0 bytes
> readahead02    2  TPASS  :  readahead saved some I/O
> readahead02    0  TINFO  :  cache can hold at least: 75140 kB
> readahead02    0  TINFO  :  read_testfile(0) used cache: 65592 kB
> readahead02    0  TINFO  :  read_testfile(1) used cache: 65592 kB
> readahead02    3  TPASS  :  using cache as expected
> 
> 
> With my patch for Btrfs:
> 
> readahead02    0  TINFO  :  creating test file of size: 67108864
> readahead02    0  TINFO  :  Looking for device in
> '/sys/fs/btrfs/60cb755a-ae5c-4059-ae0f-7a11ee9d0e9d/devices/'
> readahead02    0  TINFO  :  Reading
> /sys/fs/btrfs/60cb755a-ae5c-4059-ae0f-7a11ee9d0e9d/devices//sda2/../queue/read_ahead_kb
> readahead02    0  TINFO  :  max readahead size is: 524288
> readahead02    0  TINFO  :  read_testfile(0)
> readahead02    0  TINFO  :  Looking for device in
> '/sys/fs/btrfs/60cb755a-ae5c-4059-ae0f-7a11ee9d0e9d/devices/'
> readahead02    0  TINFO  :  Reading
> /sys/fs/btrfs/60cb755a-ae5c-4059-ae0f-7a11ee9d0e9d/devices//sda2/../queue/read_ahead_kb
> readahead02    0  TINFO  :  max readahead size is: 524288
> readahead02    0  TINFO  :  read_testfile(1)
> readahead02    0  TINFO  :  Looking for device in
> '/sys/fs/btrfs/60cb755a-ae5c-4059-ae0f-7a11ee9d0e9d/devices/'
> readahead02    0  TINFO  :  Reading
> /sys/fs/btrfs/60cb755a-ae5c-4059-ae0f-7a11ee9d0e9d/devices//sda2/../queue/read_ahead_kb
> readahead02    0  TINFO  :  max readahead size is: 524288
> readahead02    1  TPASS  :  expected ret success - returned value = 0
> readahead02    2  TPASS  :  offset is still at 0 as expected
> readahead02    0  TINFO  :  read_testfile(0) took: 117244 usec
> readahead02    0  TINFO  :  read_testfile(1) took: 34219 usec
> readahead02    0  TINFO  :  read_testfile(0) read: 67190784 bytes
> readahead02    0  TINFO  :  read_testfile(1) read: 0 bytes
> readahead02    3  TPASS  :  readahead saved some I/O
> readahead02    0  TINFO  :  cache can hold at least: 67068 kB
> readahead02    0  TINFO  :  read_testfile(0) used cache: 65592 kB
> readahead02    0  TINFO  :  read_testfile(1) used cache: 65592 kB
> readahead02    4  TPASS  :  using cache as expected
> 
> 
> 
> Btw the test should probably be cleaned up a bit, there are couple of places
> that could use SAFE_FILE_SCANF/PRINTF and it should also use monotonic timers
> instead of gettimeofday as well...
> 
> --
> Cyril Hrubis
> chrubis@suse.cz
>