Mellanox CX5-cards can not use VLAN-tagged traffic on SFR

[Robert Blomberg]

Hi,

was testing Mellanox CX5 cards with VLAN-tagged traffic in yaml-file.

So config looked like this:

[root@lab3-trex05 v2.51]# cat mlx.cfg
### Config file generated by dpdk_setup_ports.py ###

- version: 2
  interfaces: ['03:00.1', '02:00.1']
  port_info:
      - dest_mac: 50:6b:4b:d3:db:8d # MAC OF LOOPBACK TO IT'S DUAL INTERFACE
        src_mac:  50:6b:4b:c4:0c:51
        vlan : 1
      - dest_mac: 50:6b:4b:c4:0c:51 # MAC OF LOOPBACK TO IT'S DUAL INTERFACE
        src_mac:  50:6b:4b:d3:db:8d
        vlan : 2
  platform:
      master_thread_id: 0
      latency_thread_id: 27
      dual_if:
        - socket: 0
          threads: [1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26]

And I did not get any vlan tagged traffic when using version 2.51 with command: ./t-rex-64 --cfg mlx.cfg -f cap2/sfr4.yaml -m 400 -c 22

But when I tested with X710-cards VLAN-tagged traffic worked. Known issue?

Used CentOS 7.4 and OFED MLNX_OFED_LINUX-4.4-2.0.7.0-rhel7.5-x86_64.tgz.

Thanks.

/Robert Blomberg

[hanoh haim]

Could you test ASTF with latency ?

I expect only the latency streams not to work in this case.

Hanoh

--
You received this message because you are subscribed to the Google Groups "TRex Traffic Generator" group.
To unsubscribe from this group and stop receiving emails from it, send an email to trex-tgn+u...@googlegroups.com.
To post to this group, send email to trex...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/trex-tgn/7ab2aba1-8542-487a-9392-ab45b81334a3%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

--

Hanoh

Sent from my iPhone

[Robert Blomberg]

Hi,

I tested ASTF and used flag "-l 1000" if that was what you meant.

Only got VLAN 1 on VLAN-tag and some VLAN 2 on the latency packets. Very small portion of untagged traffic when testing with or without latency packets.

When using "-l 1000" on STF about 50% was tagged with VLAN 1 and VLAN 2 from respective direction.

So in other words, 25% VLAN 1, 25% VLAN 2 and 50% VLAN 0 with STF-traffic mix, and almost only VLAN 1 when using ASTF with latency packets and very small portion of VLAN 0 and VLAN 2.

/Robert

[hanoh haim]

Hi,

In STF there is a special DPDK API usage that probably does not work for Mellanox.

In ASTF this API is used only with latency streams.

Is it a new issue?

Thanks

Hanoh

To view this discussion on the web visit https://groups.google.com/d/msgid/trex-tgn/e7ef7a7f-fcf0-4ed1-bcfb-b45cfb6c006a%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[Robert Blomberg]

Hi,

I don't know if this is a new issue. We lately have started to use VLAN-tagged traffic in some of our performance test for our DUTs with MLX CX5 - cards.

Should we test previous releases of trex to verify this?

/Robert

[hanoh haim]

No need, I will look into this

To view this discussion on the web visit https://groups.google.com/d/msgid/trex-tgn/761bd885-a698-43e5-91e0-06cc6c2a321c%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[hanoh haim]

Hi, Again Mellanox is a special driver :-)

Check out this quick fix. It seems that only Mellanox driver requires the explicit enablement  of this feature and has some performance effect if it enabled (vector optimization is not used)

I've enabled it in STF/ASTF by default, didn't check regression. 

https://github.com/hhaim/trex-core/commit/04a3c1e0188db6d177834416d85dc9fd1b1eb331

Will run our regression to verify it on other setups (I might enable this only for Mellanox if there would be a problem)

thanks

Hanoh

[Robert Blomberg]

Hi,

the fix is working.

Any percentage how much more "bad" the performance get?

We will use your workaround for now, but guess this fix will be pushed in upcoming release.

Thanks for fast feedback!

/Robert

[hanoh haim]

I've pushed it into a test branch. It will be pushed to master 

I don't think there would be a big performance effect with STF/ASTF  -- but our tests will report that 

To view this discussion on the web visit https://groups.google.com/d/msgid/trex-tgn/f0f7487e-7da7-49ee-9ad9-8a38cd128a5c%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.

[Paweł Małachowski]

W dniu wtorek, 5 marca 2019 09:59:10 UTC+1 użytkownik Robert Blomberg napisał:

Any percentage how much more "bad" the performance get?

Don't know about TRex impact but in our tests DPDK testpmd and in-house packet generators show 148.8Mpps->~100Mpps TX performance drop with CX5 (FW 16.22.1002) and DPDK 18.08.

The problem is that enabling DEV_TX_OFFLOAD_VLAN_INSERT disables EMPS feature, as explained in mlx5 PMD docs.

For this reason we have implemented software VLAN encapsulation enabled on mlx5-based setups. ;)

-- 

@pawmal80

[hanoh haim]

It seems that the performance is about the same with this change for STF/ASTF. STL is the same 

--

You received this message because you are subscribed to the Google Groups "TRex Traffic Generator" group.
To unsubscribe from this group and stop receiving emails from it, send an email to trex-tgn+u...@googlegroups.com.
To post to this group, send email to trex...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/trex-tgn/cf54da98-234e-4d9b-9630-a64fb58fc3ef%40googlegroups.com.

For more options, visit https://groups.google.com/d/optout.