[patch] block: make struct block_device cacheline_aligned

[patch] block: make struct block_device cacheline_aligned

[Jeff Moyer]

Hi,

When testing against a pcie ssd or a ramdisk, making the block device
structure cacheline_aligned provided a significant increase in
performance:

                                    vanilla
          	        READ                WRITE                 CPU          
  Job Name	     BW   IOPS  msec      BW   IOPS  msec   usr  sys   csw
    write1	      0      0     0  748522 187130 44864 16.34 60.65 3799440.00
     read1	 690615 172653 48602       0      0     0 13.45 61.42 4044720.00
randwrite1	      0      0     0  716406 179101 46839 29.03 52.79 3151140.00
 randread1	 683466 170866 49108       0      0     0 25.92 54.67 3081610.00
readwrite1	 377518  94379 44450  377645  94410 44450 15.49 64.32 3139240.00
   randrw1	 355815  88953 47178  355733  88933 47178 27.96 54.24 2944570.00

                                   patched
          	        READ                WRITE                 CPU          
  Job Name	     BW   IOPS  msec      BW   IOPS  msec   usr  sys   csw
    write1	      0      0     0  871355 217838 38508 17.49 42.46 1642870.00
     read1	1418560 354639 23675       0      0     0 14.96 54.75 337489.00
randwrite1	      0      0     0  736970 184242 45633 30.62 35.25 1409440.00
 randread1	1065440 266359 31544       0      0     0 32.67 43.74 255394.00
readwrite1	 657940 164484 25867  657848 164461 25867 18.54 50.55 619474.00
   randrw1	 491940 122985 34245  492014 123003 34245 34.44 41.05 418999.00

                                   %diff
          	        READ                WRITE                 CPU          
  Job Name	     BW   IOPS  msec      BW   IOPS  msec   usr  sys   csw
    write1	      0      0     0      16     16   -14 7.04 -29.99 -56.76
     read1	    105    105   -51       0      0     0 11.23 -10.86 -91.66
randwrite1	      0      0     0       0      0     0 5.48 -33.23 -55.27
 randread1	     55     55   -35       0      0     0 26.04 -19.99 -91.71
readwrite1	     74     74   -41      74     74   -41 19.69 -21.41 -80.27
   randrw1	     38     38   -27      38     38   -27 23.18 -24.32 -85.77

BW=bandwidth in KB/s
IOPS = I/Os per second
msec = # of miliseconds the run took (lower is better)
usr = % user time
sys = % system time
csw = # of context switches

The test is doing asynchronous direct I/O to the block device using 4
processes each driving a queue depth of 1024 to a different part of the
disk.  The rows, in order, are sequential write, sequential read,
random write, random read, 50% mix of sequential reads and sequential
writes, 50% mix of random reads and random writes.  The block size in
all cases is 4k.

I'd appreciate it if others could verify an increase in performance with
this patch.  Thanks to Mikulas for initially suggesting that the cache
size/alignment was relevant to performance.

Cheers,
Jeff

Signed-off-by: Jeff Moyer <jmoyer@redhat.com>

diff --git a/include/linux/fs.h b/include/linux/fs.h
index aa11047..87ce6ca 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -724,7 +724,7 @@ struct block_device {
 	int			bd_fsfreeze_count;
 	/* Mutex for freeze */
 	struct mutex		bd_fsfreeze_mutex;
-};
+} __cacheline_aligned;
 
 /*
  * Radix-tree tags, for tagging dirty and writeback pages within the pagecache

Re: [patch] block: make struct block_device cacheline_aligned

[Jeff Moyer]

Jeff Moyer <jmoyer@redhat.com> writes:

> Hi,
>
> When testing against a pcie ssd or a ramdisk, making the block device
> structure cacheline_aligned provided a significant increase in
> performance:

Self-NACK on this one.  This results in a ton of warnings:

include/linux/fs.h:727: warning: ???__section__??? attribute does not
apply to types
In file included from include/linux/debugfs.h:18,
                 from kernel/trace/trace_probe.h:28,
                 from kernel/trace/trace_kprobe.c:23:
include/linux/fs.h:727: warning: ???__section__??? attribute does not
apply to types

And that leaves me with the task of figuring out if/why this actually
helps.

Cheers,
Jeff
>

Re: [patch] block: make struct block_device cacheline_aligned

[Mikulas Patocka]

On Wed, 19 Sep 2012, Jeff Moyer wrote:

> Jeff Moyer <jmoyer@redhat.com> writes:
> 
> > Hi,
> >
> > When testing against a pcie ssd or a ramdisk, making the block device
> > structure cacheline_aligned provided a significant increase in
> > performance:
> 
> Self-NACK on this one.  This results in a ton of warnings:
> 
> include/linux/fs.h:727: warning: ???__section__??? attribute does not
> apply to types
> In file included from include/linux/debugfs.h:18,
>                  from kernel/trace/trace_probe.h:28,
>                  from kernel/trace/trace_kprobe.c:23:
> include/linux/fs.h:727: warning: ???__section__??? attribute does not
> apply to types
> 
> And that leaves me with the task of figuring out if/why this actually
> helps.
> 
> Cheers,
> Jeff

Hi

Use ____cacheline_aligned instead of __cacheline_aligned

Mikulas

Re: [patch] block: make struct block_device cacheline_aligned

[Jeff Moyer]

Mikulas Patocka <mpatocka@redhat.com> writes:

> On Wed, 19 Sep 2012, Jeff Moyer wrote:
>
>> Jeff Moyer <jmoyer@redhat.com> writes:
>> 
>> > Hi,
>> >
>> > When testing against a pcie ssd or a ramdisk, making the block device
>> > structure cacheline_aligned provided a significant increase in
>> > performance:
>> 
>> Self-NACK on this one.  This results in a ton of warnings:
>> 
>> include/linux/fs.h:727: warning: ???__section__??? attribute does not
>> apply to types
>> In file included from include/linux/debugfs.h:18,
>>                  from kernel/trace/trace_probe.h:28,
>>                  from kernel/trace/trace_kprobe.c:23:
>> include/linux/fs.h:727: warning: ???__section__??? attribute does not
>> apply to types
>> 
>> And that leaves me with the task of figuring out if/why this actually
>> helps.
>> 
>> Cheers,
>> Jeff
>
> Hi
>
> Use ____cacheline_aligned instead of __cacheline_aligned

struct block_device is allocated as part of the bdev_inode:

struct bdev_inode {
        struct block_device bdev;
        struct inode vfs_inode;
};

The bdev_inode is allocated from the bdev_cachep, which uses
SLAB_HWCACHE_ALIGN.  So, in theory, this should already be aligned.

-Jeff

Re: [patch] block: make struct block_device cacheline_aligned

[Mikulas Patocka]

On Wed, 19 Sep 2012, Jeff Moyer wrote:

> Mikulas Patocka <mpatocka@redhat.com> writes:
> 
> > On Wed, 19 Sep 2012, Jeff Moyer wrote:
> >
> >> Jeff Moyer <jmoyer@redhat.com> writes:
> >> 
> >> > Hi,
> >> >
> >> > When testing against a pcie ssd or a ramdisk, making the block device
> >> > structure cacheline_aligned provided a significant increase in
> >> > performance:
> >> 
> >> Self-NACK on this one.  This results in a ton of warnings:
> >> 
> >> include/linux/fs.h:727: warning: ???__section__??? attribute does not
> >> apply to types
> >> In file included from include/linux/debugfs.h:18,
> >>                  from kernel/trace/trace_probe.h:28,
> >>                  from kernel/trace/trace_kprobe.c:23:
> >> include/linux/fs.h:727: warning: ???__section__??? attribute does not
> >> apply to types
> >> 
> >> And that leaves me with the task of figuring out if/why this actually
> >> helps.
> >> 
> >> Cheers,
> >> Jeff
> >
> > Hi
> >
> > Use ____cacheline_aligned instead of __cacheline_aligned
> 
> struct block_device is allocated as part of the bdev_inode:
> 
> struct bdev_inode {
>         struct block_device bdev;
>         struct inode vfs_inode;
> };
> 
> The bdev_inode is allocated from the bdev_cachep, which uses
> SLAB_HWCACHE_ALIGN.  So, in theory, this should already be aligned.
> 
> -Jeff

The purpose here is to align vfs_inode. If you add alignment to bdev, 
vfs_inode would be aligned (because bdev size would be aligned to 
cacheline boundary). Or you can add the alignment to vfs_inode, it would 
have the same effect.

Mikulas

Re: [patch] block: make struct block_device cacheline_aligned

[Jeff Moyer]

Mikulas Patocka <mpatocka@redhat.com> writes:

> On Wed, 19 Sep 2012, Jeff Moyer wrote:
>
>> Mikulas Patocka <mpatocka@redhat.com> writes:
>> 
>> > On Wed, 19 Sep 2012, Jeff Moyer wrote:
>> >
>> >> Jeff Moyer <jmoyer@redhat.com> writes:
>> >> 
>> >> > Hi,
>> >> >
>> >> > When testing against a pcie ssd or a ramdisk, making the block device
>> >> > structure cacheline_aligned provided a significant increase in
>> >> > performance:
>> >> 
>> >> Self-NACK on this one.  This results in a ton of warnings:
>> >> 
>> >> include/linux/fs.h:727: warning: ???__section__??? attribute does not
>> >> apply to types
>> >> In file included from include/linux/debugfs.h:18,
>> >>                  from kernel/trace/trace_probe.h:28,
>> >>                  from kernel/trace/trace_kprobe.c:23:
>> >> include/linux/fs.h:727: warning: ???__section__??? attribute does not
>> >> apply to types
>> >> 
>> >> And that leaves me with the task of figuring out if/why this actually
>> >> helps.
>> >> 
>> >> Cheers,
>> >> Jeff
>> >
>> > Hi
>> >
>> > Use ____cacheline_aligned instead of __cacheline_aligned
>> 
>> struct block_device is allocated as part of the bdev_inode:
>> 
>> struct bdev_inode {
>>         struct block_device bdev;
>>         struct inode vfs_inode;
>> };
>> 
>> The bdev_inode is allocated from the bdev_cachep, which uses
>> SLAB_HWCACHE_ALIGN.  So, in theory, this should already be aligned.
>> 
>> -Jeff
>
> The purpose here is to align vfs_inode.

I'm not sure I understand what you're saying.  When you say, "The
purpose here," do you mean the purpose in the existing code or the
purpose of our changes?  The existing code seems to want to align the
struct block_device, so I assume you mean we should instead align the
vfs_inode.

> If you add alignment to bdev, vfs_inode would be aligned (because bdev
> size would be aligned to cacheline boundary).

ITYM because the bdev size *is* aligned to a cacheline boundary (the
size is 256 in my kernel, and the cache line alignment for this cpu is
64).  But, since the entire structure is already aligned by the slab
allocator, I don't see how adding ____cacheline_aligned would change
anything.

> Or you can add the alignment to vfs_inode, it would have the same
> effect.

Well, I tried the suggestion by Richard to swap the fields in the
bdev_inode, and it did not result in a huge performance gain:

                                   %diff
          	        READ                WRITE                 CPU          
  Job Name	     BW   IOPS  msec      BW   IOPS  msec   usr  sys   csw
    write1	      0      0     0       9      9    -8 0.00 0.00 -17.18
     read1	      6      6    -6       0      0     0 5.87 0.00 -19.20
randwrite1	      0      0     0       0      0     0 0.00 0.00 -15.46
 randread1	      5      5    -5       0      0     0 0.00 0.00 -13.69
readwrite1	      0      0     0       0      0     0 0.00 0.00   7.83
   randrw1	      5      5    -5       5      5    -5 0.00 0.00 -12.29

I can try adding the ____cacheline_aligned to the vfs_inode inside of
the bdev_inode if you like.  Any other ideas?

Cheers,
Jeff

Re: [patch] block: make struct block_device cacheline_aligned

[Mikulas Patocka]

On Wed, 19 Sep 2012, Jeff Moyer wrote:

> Mikulas Patocka <mpatocka@redhat.com> writes:
> 
> > On Wed, 19 Sep 2012, Jeff Moyer wrote:
> >
> >> Mikulas Patocka <mpatocka@redhat.com> writes:
> >> 
> >> > On Wed, 19 Sep 2012, Jeff Moyer wrote:
> >> >
> >> >> Jeff Moyer <jmoyer@redhat.com> writes:
> >> >> 
> >> >> > Hi,
> >> >> >
> >> >> > When testing against a pcie ssd or a ramdisk, making the block device
> >> >> > structure cacheline_aligned provided a significant increase in
> >> >> > performance:
> >> >> 
> >> >> Self-NACK on this one.  This results in a ton of warnings:
> >> >> 
> >> >> include/linux/fs.h:727: warning: ???__section__??? attribute does not
> >> >> apply to types
> >> >> In file included from include/linux/debugfs.h:18,
> >> >>                  from kernel/trace/trace_probe.h:28,
> >> >>                  from kernel/trace/trace_kprobe.c:23:
> >> >> include/linux/fs.h:727: warning: ???__section__??? attribute does not
> >> >> apply to types
> >> >> 
> >> >> And that leaves me with the task of figuring out if/why this actually
> >> >> helps.
> >> >> 
> >> >> Cheers,
> >> >> Jeff
> >> >
> >> > Hi
> >> >
> >> > Use ____cacheline_aligned instead of __cacheline_aligned
> >> 
> >> struct block_device is allocated as part of the bdev_inode:
> >> 
> >> struct bdev_inode {
> >>         struct block_device bdev;
> >>         struct inode vfs_inode;
> >> };
> >> 
> >> The bdev_inode is allocated from the bdev_cachep, which uses
> >> SLAB_HWCACHE_ALIGN.  So, in theory, this should already be aligned.
> >> 
> >> -Jeff
> >
> > The purpose here is to align vfs_inode.
> 
> I'm not sure I understand what you're saying.  When you say, "The
> purpose here," do you mean the purpose in the existing code or the
> purpose of our changes?  The existing code seems to want to align the
> struct block_device, so I assume you mean we should instead align the
> vfs_inode.
> 
> > If you add alignment to bdev, vfs_inode would be aligned (because bdev
> > size would be aligned to cacheline boundary).
> 
> ITYM because the bdev size *is* aligned to a cacheline boundary (the
> size is 256 in my kernel, and the cache line alignment for this cpu is
> 64).  But, since the entire structure is already aligned by the slab
> allocator, I don't see how adding ____cacheline_aligned would change
> anything.
> 
> > Or you can add the alignment to vfs_inode, it would have the same
> > effect.
> 
> Well, I tried the suggestion by Richard to swap the fields in the
> bdev_inode, and it did not result in a huge performance gain:
> 
>                                    %diff
>           	        READ                WRITE                 CPU          
>   Job Name	     BW   IOPS  msec      BW   IOPS  msec   usr  sys   csw
>     write1	      0      0     0       9      9    -8 0.00 0.00 -17.18
>      read1	      6      6    -6       0      0     0 5.87 0.00 -19.20
> randwrite1	      0      0     0       0      0     0 0.00 0.00 -15.46
>  randread1	      5      5    -5       0      0     0 0.00 0.00 -13.69
> readwrite1	      0      0     0       0      0     0 0.00 0.00   7.83
>    randrw1	      5      5    -5       5      5    -5 0.00 0.00 -12.29
> 
> I can try adding the ____cacheline_aligned to the vfs_inode inside of
> the bdev_inode if you like.  Any other ideas?
> 
> Cheers,
> Jeff

I added ____cacheline_aligned - the results are this:
3.5.4 (without ____cacheline_aligned):	60.0630s, stdev 0.9087
3.5.4, ____cacheline_aligned:		43.1266s, stdev 0.8916
3.5.4, ____cacheline_aligned, patch 1:	42.7746s, stdev 0.3491
3.5.4, ____cacheline_aligned, patch 2:	45.1152s, stdev 0.8554
3.5.4, ____cacheline_aligned, patch 3:	43.2462s, stdev 0.1946
3.5.4, ____cacheline_aligned, patch 4:	42.8494s, stdev 0.2387

--- so, cacheline_aligned makes the results consistent. Patch 4 looks 
slighly better than patch 3. What are your results with cacheline_aligned?

Mikulas

Re: [patch] block: make struct block_device cacheline_aligned

[Richard Kennedy]

On 19/09/12 15:59, Jeff Moyer wrote:
> Jeff Moyer <jmoyer@redhat.com> writes:
> 
>> Hi,
>>
>> When testing against a pcie ssd or a ramdisk, making the block device
>> structure cacheline_aligned provided a significant increase in
>> performance:
> 
> Self-NACK on this one.  This results in a ton of warnings:
> 
> include/linux/fs.h:727: warning: ???__section__??? attribute does not
> apply to types
> In file included from include/linux/debugfs.h:18,
>                  from kernel/trace/trace_probe.h:28,
>                  from kernel/trace/trace_kprobe.c:23:
> include/linux/fs.h:727: warning: ???__section__??? attribute does not
> apply to types
> 
> And that leaves me with the task of figuring out if/why this actually
> helps.
> 
> Cheers,
> Jeff
>>
Hi,
block_device sits in bdev_inode which is already cache line aligned as
they're allocated in a kmem_cache with HWALIGN set.
So I wonder if swapping the order of bdev & inode in bdev_inode will help?

So it becomes this :-
struct bdev_inode {
	struct inode vfs_inode;
	struct block_device bdev;
};


regards
Richard