Re: [dm-devel] [PATCH 1/1] block: Convert hd_struct in_flight from atomic to percpu

[Brian King <brking@linux.vnet.ibm.com>]

On 07/01/2017 11:43 AM, Jens Axboe wrote:
> Now:
> 
> http://git.kernel.dk/cgit/linux-block/commit/?h=mq-inflight&id=87f73ef2b9edb6834001df8f7cb48c7a116e8cd3
> 
>> And updated the branch here:
>>
>> http://git.kernel.dk/cgit/linux-block/log/?h=mq-inflight
>>
>> to include that, and be based on top of for-4.13/block. If you prefer just
>> pulling a branch, pull:
>>
>> git://git.kernel.dk/linux-block mq-inflight
> 
> I've tested it, and made a few adjustments and fixes. The branches are
> the same, just rebased. Works fine for me, even with partitions now.
> 
> Let me know how it works for you.

This is looking very nice. This is what I'm seeing now:

1 null_blk device / 80 submit_queues / fio numjobs=80
9.4M IOPs

80 null_blk devices / 1 submit_queue per null_blk / fio numjobs=1 (1 per null_blk)
9.3M IOPs

1 null_blk device / 160 submit_queues / fio numjobs=160
9.8M IOPs

160 null_blk devices / 1 submit_queue per null_blk / fio numjobs=1 (1 per null_blk)
8.9M IOPs

What is queued in mq-inflight looks to resolve the issue I've been seeing.

Thanks!!!

Brian

-- 
Brian King
Power Linux I/O
IBM Linux Technology Center