Add payload threshold for using uTP, and show work

[carver]

We don't appear to explicitly say the threshold for switching from inline content to uTP delivery anywhere, yet. It's referenced in hive, so we should presumably actually agree on the explicit limit: https://github.com/ethereum/hive/blob/8e3ffff6ea74d20fc073f9a78baa574e7e688856/simulators/portal/src/suites/history/interop.rs#L15

[KolbyML]

It is referenced from hive as I was writing tests which required it. Anyways going forward I should updated the tests to use MAX_PORTAL_CONTENT_PAYLOAD_SIZE which is defined in ethereum/trin ethportal-api, so it is always the right number going forward into the future. As with the native Discv5 uTP proposal (instead of tunneling uTP through TalkReq), the max size would get bigger.

Anyways I think this proposal to formalize this so future implementer's have an easier time is a good thing so...

[carver]

MAX_PORTAL_CONTENT_PAYLOAD_SIZE which is defined in ethereum/trin ethportal-api

Yeah, I nearly linked to it in these specs, but decided it was probably not best practice to imply that any particular implementation owned the canonical definition of the max payload size.

[kdeme]

Great for adding this in the specifications, but I believe the value should be 1175 bytes.

That's the value that gets calculated by Fluffy at least, and IIRC this was at some point in the past also looked at as reference by Trin, but perhaps it was not taken over exactly or got altered over time.

Calculation in Fluffy happens here: https://github.com/status-im/nimbus-eth1/blob/master/fluffy/network/wire/portal_protocol.nim#L123 & https://github.com/status-im/nimbus-eth1/blob/master/fluffy/network/wire/portal_protocol.nim#L404-L406

I don't have an exact test for this, but I do remember verifying this byte per byte on a encoded talkresp message.
I will make the consts more clear and can add a test for this also to show.

[kdeme]

I have moved the calculation for talkreq overhead to our discv5 module and added tests there: status-im/nim-eth@aa1e738

So I'm pretty certain that these are the correct values (at least with our discv5 impl., but that shouldn't differ really).

[carver]

It looks like this test is for the max size of the Talk Response payload, but the relevant payloads for a Content message are transported in a Talk Request. Does that make a difference?

[kdeme]

The content message adds just two additional bytes of overhead, so those need to be added still:

https://github.com/status-im/nimbus-eth1/blob/master/fluffy/network/wire/portal_protocol.nim#L391

You can also see that in our test vectors: https://github.com/ethereum/portal-network-specs/blob/master/portal-wire-test-vectors.md#content-response---content-payload

[carver]

Ok, I'll update the spec to note that we're talking about the size of the payload inside the message, and point out those extra 2 bytes.

I don't see a test in trin for this content size max, so I'll take your values here :)

[carver]

Wait, but it looks like the message overhead is different for the different scenarios, like find nodes vs find content, so maybe I'll just ignore the 2 bytes here, and look at the total encoded payload size

[kdeme]

Yes, it will be different for each of the wire protocol messages. But I guess the most important one here is the one that differentiates between sending content over the protocol response or over uTP

[pipermerriam]

Overall, 👍 .

nitpick: The choice of rust for the code snippet seems a little counter-intuitive to me in terms of readability. Python or a similar loosely typed language that allows us to blindly deal with integers would probably have less extra notation.