[dpdk-dev] [Qemu-devel] [PATCH 1/2] vhost: enable any layout feature

[Maxime Coquelin]

On 10/10/2016 04:42 PM, Yuanhan Liu wrote:
> On Mon, Oct 10, 2016 at 02:40:44PM +0200, Maxime Coquelin wrote:
>>>>> At that time, a packet always use 2 descs. Since indirect desc is
>>>>> enabled (by default) now, the assumption is not true then. What's
>>>>> worse, it might even slow things a bit down. That should also be
>>>>> part of the reason why performance is slightly worse than before.
>>>>>
>>>>> 	--yliu
>>>>
>>>> I'm not sure I get what you are saying
>>>>
>>>>> commit 1d41d77cf81c448c1b09e1e859bfd300e2054a98
>>>>> Author: Yuanhan Liu <yuanhan.liu at linux.intel.com>
>>>>> Date:   Mon May 2 17:46:17 2016 -0700
>>>>>
>>>>>    vhost: optimize dequeue for small packets
>>>>>
>>>>>    A virtio driver normally uses at least 2 desc buffers for Tx: the
>>>>>    first for storing the header, and the others for storing the data.
>>>>>
>>>>>    Therefore, we could fetch the first data desc buf before the main
>>>>>    loop, and do the copy first before the check of "are we done yet?".
>>>>>    This could save one check for small packets that just have one data
>>>>>    desc buffer and need one mbuf to store it.
>>>>>
>>>>>    Signed-off-by: Yuanhan Liu <yuanhan.liu at linux.intel.com>
>>>>>    Acked-by: Huawei Xie <huawei.xie at intel.com>
>>>>>    Tested-by: Rich Lane <rich.lane at bigswitch.com>
>>>>
>>>> This fast-paths the 2-descriptors format but it's not active
>>>> for indirect descriptors. Is this what you mean?
>>>
>>> Yes. It's also not active when ANY_LAYOUT is actually turned on.
>>>> Should be a simple matter to apply this optimization for indirect.
>>>
>>> Might be.
>>
>> If I understand the code correctly, indirect descs also benefit from this
>> optimization, or am I missing something?
>
> Aha..., you are right!

The interesting thing is that the patch I send on Thursday that removes
header access when no offload has been negotiated[0] seems to reduce
almost to zero the performance seen with indirect descriptors enabled.
I see this with 64 bytes packets using testpmd on both ends.

When I did the patch, I would have expected the same gain with both
modes, whereas I measured +1% for direct and +4% for indirect.

Maxime
[0]: http://dpdk.org/dev/patchwork/patch/16423/