B. 🥊 Decentralized Hole Punching

[p-shahi]

eta: 2023Q2

Why: P2P networks can have a combination of both public and private nodes. While private nodes can dial nodes on the public Internet, they are unreachable from the outside as they are behind a NAT or a firewall. We need a mechanism to dial them. A previous DHT crawl found that almost 63% of the network was undialable.

Goal: Implement Decentralized Hole Punching in js-libp2p and bring it to parity with the Go and Rust implementations.

• Autonat support #1005
• Implement Circuit Relay v2 in JS #1029
• Add support for QUIC #1459
• Use DCUtR to synchronize hole punching #1460

[momack2]

can we clarify that this is for js-ipfs in the title please?

[p-shahi]

@momack2 Any reason to include js-ipfs specifically? This work is not progressing with it in mind. We will need it for Helia but I would still not include any project in the title since it's better to remain product agnostic

[momack2]

sorry - I think I meant "js-libp2p" - the reason is because each of these cards shows up with out the repo title in starmaps - so otherwise you have many "holepunching" milesotnes without knowing they are for different libp2p implementations

[SgtPooki]

@momack2 we could include the repo-name on milestone cards in starmap to help mitigate this problem?

[p-shahi]

Didn't see the replies above but I support adding the repo-name to the milestone cards

[momack2]

Yeah, I think that would help a lot

…

On Tue, May 9, 2023 at 2:39 AM Prithvi Shahi ***@***.***> wrote: Didn't see the replies above but I support adding the repo-name to the milestone cards — Reply to this email directly, view it on GitHub <#1461 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAEXAF7RQHPNSIA76WOYPQLXFHRALANCNFSM6AAAAAARKTHWP4> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[2color]

@p-shahi Is this still in progress? It was my understanding that this is already implemented in js-libp2p?

[SgtPooki]

JS Colo 2024 discussion

We don't do hole-punching in TCP, hole-punching is much more successful with UDP (timing is more reliable)

• TCP: ~30% success
• UDP: ~70% success
• Ref: https://youtu.be/bzL7Y1wYth8?feature=shared&t=1157

Things left to do:

1. Incorporate QUIC transport
2. Implement missing parts of DCUtR (the half after unilateral connection upgrade)