Re: [Qemu-devel] [PATCHv4 0/9] buffer_is_zero / migration optimizations

[Peter Lieven]

Am 22.03.2013 18:25, schrieb Paolo Bonzini:
> Il 22/03/2013 13:46, Peter Lieven ha scritto:
>> this is v4 of my patch series with various optimizations in
>> zero buffer checking and migration tweaks.
>>
>> thanks especially to Eric Blake for reviewing.
>>
>> v4:
>> - do not inline buffer_find_nonzero_offset()
>> - inline can_usebuffer_find_nonzero_offset() correctly
>> - readd asserts in buffer_find_nonzero_offset() as profiling
>>   shows they do not hurt.
>> - change last occurences of scalar 8 by 
>>   BUFFER_FIND_NONZERO_OFFSET_UNROLL_FACTOR
>> - avoid deferencing p already in patch 5 where we
>>   know that the page (p) is zero
>> - explicitly set bytes_sent = 0 if we skip a zero page.
>>   bytes_sent was 0 before, but it was not obvious.
>> - add accounting information for skipped zero pages
>> - fix errors reported by checkpatch.pl
>>
>> v3:
>> - remove asserts, inline functions and add a check
>>   function if buffer_find_nonzero_offset() can be used.
>> - use above check function in buffer_is_zero() and
>>   find_next_bit().
>> - use buffer_is_nonzero_offset() directly to find
>>   zero pages. we know that all requirements are met
>>   for memory pages.
>> - fix C89 violation in buffer_is_zero().
>> - avoid derefencing p in ram_save_block() if we already
>>   know the page is zero.
>> - fix initialization of last_offset in reset_ram_globals().
>> - avoid skipping pages with offset == 0 in bulk stage in
>>   migration_bitmap_find_and_reset_dirty().
>> - compared to v1 check for zero pages also after bulk
>>   ram migration as there are guests (e.g. Windows) which
>>   zero out large amount of memory while running.
>>
>> v2:
>> - fix description, add trivial zero check and add asserts 
>>   to buffer_find_nonzero_offset.
>> - add a constant for the unroll factor of buffer_find_nonzero_offset
>> - replace is_dup_page() by buffer_is_zero()
>> - added test results to xbzrle patch
>> - optimize descriptions
>>
>> Peter Lieven (9):
>>   move vector definitions to qemu-common.h
>>   cutils: add a function to find non-zero content in a buffer
>>   buffer_is_zero: use vector optimizations if possible
>>   bitops: use vector algorithm to optimize find_next_bit()
>>   migration: search for zero instead of dup pages
>>   migration: add an indicator for bulk state of ram migration
>>   migration: do not sent zero pages in bulk stage
>>   migration: do not search dirty pages in bulk stage
>>   migration: use XBZRLE only after bulk stage
>>
>>  arch_init.c                   |   74 
>> +++++++++++++++++++----------------------
>>  hmp.c                         |    2 ++
>>  include/migration/migration.h |    2 ++
>>  include/qemu-common.h         |   37 +++++++++++++++++++++
>>  migration.c                   |    3 +-
>>  qapi-schema.json              |    6 ++--
>>  qmp-commands.hx               |    3 +-
>>  util/bitops.c                 |   24 +++++++++++--
>>  util/cutils.c                 |   50 ++++++++++++++++++++++++++++
>>  9 files changed, 155 insertions(+), 46 deletions(-)
>>
> I think patch 4 is a bit overengineered.  I would prefer the simple
> patch you had using three/four non-vectorized accesses.  The setup cost
> of the vectorized buffer_is_zero is quite high, and 64 bits are just
> 256k RAM; if the host doesn't touch 256k RAM, it will incur the overhead.
I think you are right. I was a little to eager to utilize 
buffer_find_nonzero_offset()
as much as possible. The performance gain by unrolling was impressive enough.
The gain by the vector functions is not that big that it would justify a 
possible
slow down by the high setup costs. My testings revealed that in most cases 
buffer_find_nonzero_offset()
returns 0 or a big offset. All the 0 return values would have increased setup 
costs with
the vectorized version of patch 4.

>
> I would prefer some more benchmarking for patch 5, but it looks ok.
What would you like to see? Statistics how many pages of a real system
are not zero, but zero in the first sizeof(long) bytes?

>
> The rest are fine, thanks!
Thank you for reviewing. If we are done with this patches I will continue with
the block migration optimizations next week.

Peter