Re: [PATCH 2/2] blk-plug: don't flush nested plug lists

[Jeff Moyer <jmoyer@redhat.com>]

Jens Axboe <axboe@kernel.dk> writes:

>> Comments would be greatly appreciated.
>
> It's hard to argue with the increased merging for your case. The task
> plugs did originally work like you changed them to, not flushing until
> the outermost plug was flushed. Unfortunately I don't quite remember
> why I changed them, will have to do a bit of digging to refresh my
> memory.

Let me know what you dig up.  

> For cases where we don't do any merging (like nvme), we always want to
> flush. Well almost, if we start do utilize the batched submission,
> then the plug would still potentially help (just for other reasons
> than merging).

It's never straight-forward.  :)

> And agree with Ming, this can be cleaned up substantially. I'd also

Let me know if you have any issues with the v2 posting.

> like to see some test results from the other end of the spectrum. Your
> posted cased is clearly based case (we missed tons of merging, now we
> don't), I'd like to see a normal case and a worst case result as well
> so we have an idea of what this would do to latencies.

As for other benchmark numbers, this work was inspired by two reports of
lower performance after converting virtio_blk to use blk-mq in RHEL.
The workloads were both synthetic, though.  I'd be happy to run a
battery of tests on the patch set, but if there is anything specific you
want to see (besides a workload that will have no merges), let me know.

I guess I should also mention that another solution to the problem might
be the mythical mq I/O scheduler.  :)

Cheers,
Jeff