Performance in stmmac

Hi to all,

I finished implementing the multi-queue / multi-channel in stmmac and I am now
making some tests.

The mechanism is working properly, but the performance could be better
(~440Mb/s) :). For measuring I am using iperf:

synopsys@pt02ipk1:~$ iperf -c 192.168.0.3 --port 5001 -t 20 -i 5
------------------------------------------------------------
Client connecting to 192.168.0.3, TCP port 5001
TCP window size: 85.0 KByte (default)
------------------------------------------------------------
[  3] local 192.168.0.2 port 34272 connected with 192.168.0.3 port 5001
[ ID] Interval       Transfer     Bandwidth
[  3]  0.0- 5.0 sec   260 MBytes   436 Mbits/sec
[  3]  5.0-10.0 sec   262 MBytes   440 Mbits/sec
[  3] 10.0-15.0 sec   262 MBytes   440 Mbits/sec
[  3] 15.0-20.0 sec   257 MBytes   431 Mbits/sec
[  3]  0.0-20.0 sec  1.02 GBytes   437 Mbits/sec

I am seeing tx being performed in all queues, but rx is just being routed to
queue 0. The reason why only queue 0 is processing the packets is because I am
using priority tagged routing for now and iperf is not capable of producing
priority tagged traffic.

I am sending you this e-mail in order to know if you could suggest me a tool to
generate priority tagged traffic and if you have experience in tweaking stmmac
to improve performance.

Thanks,
Joao

From: Joao Pinto <redacted>
Date: Tue, 7 Feb 2017 19:33:52 +0000

I am seeing tx being performed in all queues, but rx is just being routed to
queue 0. The reason why only queue 0 is processing the packets is because I am
using priority tagged routing for now and iperf is not capable of producing
priority tagged traffic.

I am sending you this e-mail in order to know if you could suggest me a tool to
generate priority tagged traffic and if you have experience in tweaking stmmac
to improve performance.

Multi-queue is only going to be effective on RX if you can do proper hashed
flow separation.  That's the entire performance point of the feature.

If the chip cannot do this, your performance will be thusly limited.

Hi Joao

On 2/7/2017 8:33 PM, Joao Pinto wrote:

I am seeing tx being performed in all queues, but rx is just being routed to
queue 0. The reason why only queue 0 is processing the packets is because I am
using priority tagged routing for now and iperf is not capable of producing
priority tagged traffic.

I wonder if you tried the multi-queue channels & AV/B stack.

Regards
Peppe