Re: [PATCH v8 bpf-next 00/14] mvneta: introduce XDP multi-buffer support

From: Daniel Borkmann <daniel@iogearbox.net>
To: Lorenzo Bianconi <lorenzo.bianconi@redhat.com>,
	Magnus Karlsson <magnus.karlsson@gmail.com>
Cc: Lorenzo Bianconi <lorenzo@kernel.org>, bpf <bpf@vger.kernel.org>,
	Network Development <netdev@vger.kernel.org>,
	"David S. Miller" <davem@davemloft.net>,
	Jakub Kicinski <kuba@kernel.org>,
	Alexei Starovoitov <ast@kernel.org>,
	shayagr@amazon.com, sameehj@amazon.com,
	John Fastabend <john.fastabend@gmail.com>,
	David Ahern <dsahern@kernel.org>,
	Jesper Dangaard Brouer <brouer@redhat.com>,
	Eelco Chaudron <echaudro@redhat.com>,
	Jason Wang <jasowang@redhat.com>,
	Alexander Duyck <alexander.duyck@gmail.com>,
	Saeed Mahameed <saeed@kernel.org>,
	"Fijalkowski, Maciej" <maciej.fijalkowski@intel.com>,
	Tirthendu <tirthendu.sarkar@intel.com>
Subject: Re: [PATCH v8 bpf-next 00/14] mvneta: introduce XDP multi-buffer support
Date: Sat, 17 Apr 2021 01:00:34 +0200	[thread overview]
Message-ID: <daeb8a26-44fa-d76e-dd8e-bd4ca62ef868@iogearbox.net> (raw)
In-Reply-To: <YHoBtldcPyKNFKPv@lore-desk>

On 4/16/21 11:29 PM, Lorenzo Bianconi wrote:
>>
>> Took your patches for a test run with the AF_XDP sample xdpsock on an
>> i40e card and the throughput degradation is between 2 to 6% depending
>> on the setup and microbenchmark within xdpsock that is executed. And
>> this is without sending any multi frame packets. Just single frame
>> ones. Tirtha made changes to the i40e driver to support this new
>> interface so that is being included in the measurements.
> 
> thx for working on it. Assuming the fragmented part is only initialized/accessed
> if mb is set (so for multi frame packets), I would not expect any throughput
> degradation in the single frame scenario. Can you please share the i40e
> support added by Tirtha?

Thanks Tirtha & Magnus for adding and testing mb support for i40e, and sharing those
data points; a degradation between 2-6% when mb is not used would definitely not be
acceptable. Would be great to root-cause and debug this further with Lorenzo, there
really should be close to /zero/ additional overhead to avoid regressing existing
performance sensitive workloads like load balancers, etc once they upgrade their
kernels/drivers.

>> What performance do you see with the mvneta card? How much are we
>> willing to pay for this feature when it is not being used or can we in
>> some way selectively turn it on only when needed?
> 
> IIRC I did not get sensible throughput degradation on mvneta but I will re-run
> the tests running an updated bpf-next tree.

But compared to i40e, mvneta is also only 1-2.5 Gbps so potentially less visible,
right [0]? Either way, it's definitely good to get more data points from benchmarking
given this was lacking before for higher speed NICs in particular.

Thanks everyone,
Daniel

   [0] https://doc.dpdk.org/guides/nics/mvneta.html