packet replication and elimination using P4

[sachintiptur]

Hi P4 developers,

I am working on a internship topic related to deterministic networks. We are trying to implement packet replication and elimination as part of this. I came across p4 and started with tutorials and thought of using bmv2 simple_switch(tutorial code base and vm image) for this.

Can you please give me pointers or leads to study further and how can I implement the packet elimination and replication functionality. Also if any example code where such functionality is supported. It will be of great useful to me.

Requirement: Below is the network topology. In this, p4 switches on the edge,

1. duplicates the incoming packets from sender and sends over the two different path to the same destination

2. Eliminates the duplicate packet(if already one is received) on the receiving side
   PS: for elimination of duplicate pkt, i need to store identity of pkt(may be checksum of payload or introduce a sequence number) in a table. is it possible in p4 to write(write access from p4 program, not from control plane program) to tables while data path processing or should i consider any other way of implementing elimination algorithm?

Thanks,
Sachin

[antoninbas]

The v1model & PSA architectures offer "register" data structures that can be read & written from the data plane. These are essentially indexed arrays, and their size is constrained by the hardware executing the P4 program. Of course, when using bmv2, registers can be very large (you are only constrained by the RAM of the machine on which you are running bmv2). However, P4-programmable hardware able to process packets at Tbps speeds will typically have on-chip memory in the 10s of MBs. Therefore, identifying packets based on a signature would typically be done using Bloom filters (https://en.wikipedia.org/wiki/Bloom_filter) or a similar space-efficient data structure, the tradeoff being that false positives are possible.

There is quite a lot of scientific literature about using Bloom filters in P4-programmable devices. This university lecture also covers the topic: https://adv-net.ethz.ch/pdfs/03_stateful.pdf. You should also be able to find papers discussing specifically the topic of deduplication using P4, as this is a common function of network packet brokers.

[pcarrillom]

Hello Sachin

I am working in a very similar project and would like to know if you could make some progress. Especially in the elimination part.
Our approach also needs an alignment buffer, because both path can have different latency.

For P4 developers, I have a question:

Is there a way to implement a fixed time delay of packets for alignment?

thanks

[jfingerh]

I would recommend asking open-ended P4 related questions on this site, since then the answers will be more easily found later by others, vs. a Github issue that will likely be closed and less likely to be found in the future. https://forum.p4.org @pcarrilom