[P4-dev] P4 issues on mininet with multiple switches

Hassan Ali Khan 18100116 at lums.edu.pk
Sat May 6 17:08:34 EDT 2017


Thank you, Amedeo, for your reply.


One other issue that we are facing in mininet is that although the simple_router implementation works for one switch, it doesn't work for multiple switches. The reason, I assume, is that the table entries caters the case where the packet's next hop for the switch is the receiving host since there are no switches in between ; however, with multiple switches, new entries have to be added where we need the IPs of the switch ports, linked with each switch. Maybe that is why, our packets are not being forwarded properly.

What's your view about that?


Regards,
Hassan Ali Khan
2018-10-0116
________________________________
From: Amedeo Sapio <amedeo.sapio at polito.it>
Sent: Sunday, May 7, 2017 1:29:07 AM
To: Hassan Ali Khan
Cc: p4-dev; Haseeb Naseem
Subject: Re: [P4-dev] P4 issues on mininet with multiple switches

​Hi,
since you talk about acks, I assume you meant TCP and not UDP.
I had a similar experience using network namespaces (similar to what mininet does). The destination host was receiving all the packets but it did not send any ACK. The problem was that TCP offloading was active on the virtual interfaces, so TCP packets had the wrong checksum. You can disable it in each virtual interface used by mininet with the following 2 commands:

ethtool --offload ethX rx off tx off
ethtool -K ethX gso off

​using the name of your interface instead of ethX.

Hope this helps,
Amedeo

---
Amedeo Sapio

e-mail: amedeo.sapio at polito.it<mailto:amedeo.sapio at polito.it>
Mobile: +39 3291564391

[View Amedeo Sapio's profile on LinkedIn]<https://it.linkedin.com/pub/amedeo-sapio/50/153/25>


We have been trying to test the P4 implementation of ecmp for quite a while and have failed due to mininet causing problems with topologies containing multiple switches. For example, in our case, Host 1 is sending UDP packets and is Host 2 is receiving packets but not sending ACKS and hence , we cannot evaluate the performance of the protocol. Can you please guide us how to approach this problem?


Regards,
Hassan Ali Khan
2018-10-0116
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.p4.org/pipermail/p4-dev_lists.p4.org/attachments/20170506/f4f97242/attachment-0002.html>


More information about the P4-dev mailing list