Rerun tests using different MQ Size

[araskachoi]

• edit the queue size of gossipsub
• the value is hardcoded at 32
• try with a queue size of 128, and with a queue size of 256

NOTE: PL has also merged the pubsub tracer recently (https://github.com/libp2p/go-libp2p-pubsub/pull/227) and created a bunch of utilities for collecting, manipulating, and inspecting traces from a network (https://github.com/libp2p/go-libp2p-pubsub-tracer)

Should perform additional runs with/without tracing, and sending the dumps over so the libp2p team can inspect them.

[araskachoi]

@jasonatran the best way would be to fork go-libp2p-pubsub to make the necessary changes and pull the code into our fork of adam's code in my repo

what do you think?

[jasonatran]

I think the tracer looks promising, so we can avoid any fork of go-libp2p-pubsub for now. We can revisit this after the first attempt to test using just the new tracer implementation. They already merged go-libp2p-pubsub/pull/227 for us.

[araskachoi]

yes, but looks like that's going to require additional implementation into adam's code. for the sake of time, might be more efficient to run the tests with the changed message queue size first and then implement tracer later? or hold off on all tests until we get the tracer?

[jasonatran]

Oh yea, I do think we should fork go-libp2p-pubsub-benchmark-tools. We have a bunch of changes coming up for tracer, so I think it'll give us more agility if we fork off. I think a single run to verify that the queue size change fixes the dropped packet will be sufficient. My gut tells me it won't change the overall statistics of the gossip tests. If there is a change, we'll see it when we start testing with tracer.

[araskachoi]

will not use tracer for these tests and will move forward with using an increased messagequeuesize

https://github.com/araskachoi/go-libp2p-pubsub-benchmark-tools/blob/develop/pkg/host/host.go#L235