Bitswap peer receives duplicate blocks

[btc]

@whyrusleeping performed a bitswap test.

1. Three nodes pin a widely available file.
2. Each node receives multiple copies of the file. (In his test, each received 20. Every partner that received the three nodes' want lists sent every block on it)

Some initial discussion occurred in IRC at #ipfs. Let's open discussion around the implementation of solutions for the various inefficiencies of the current implementation. Empirical and analytical approaches are both welcome.

A couple initial idea come to mind:

• Instead of sending requests to each and every block provider (returned from the routing system), schedule requests to a small number and follow up with additional requests if necessary
• Instead of sending the local node's entire want list to each and every peer, partition the want list.

edited: with additional info from @whyrusleeping

cc
@danmane

[whyrusleeping]

Essentially what it came down to is that everyone who got out want list sent us every block on it. That's probably not good

[whyrusleeping]

Also, Since this issue is titled "Bitswap Improvements" I figure its a fair place to mention that i feel the strategies are too tightly coupled with the ledgers. I feel like the strategies should be a simpler function (or method on an object) that is passed a ledger as its state.

[btc]

@whyrusleeping

Thinking of the ledger as per-peer bitswap state... if we wish to strategize on a per-peer basis, information about the active strategy for the peer must reside within that state.

We could remove the StrategyFunc, but the ledger would still need a StrategyId. Would that assuage your concern? If yes, then I'm in agreement. If not, then we might be on different pages and I'm curious what you have in mind.

[whyrusleeping]

Im pretty sure i can mark this resolved, the latest changeset to bitswap has partially addressed this. (still not perfect, but better)