proposal for pubsub boosters

proposals/pubsub-boosters.md

@@ -0,0 +1,95 @@
+# Filecoin Network Boosters
+
+Authors: vyzo
+
+Initial PR: TBD <!-- Reference the PR first proposing this document. Oooh, self-reference! -->
+
+## Purpose &amp; impact
+#### Background &amp; intent
+
+Messages and Blocks propagate in the filecoin network using pubsub (gossipsub).
+In general, this works well enough; however it has been observed that small miners outside mainland
+China often get blocks late, which results in mining orphan blocks.
+
+This project proposes to build a small network backbone consisting of pubsub boosters that are
+directly peered with large miners in mainland China and exchanges in order to ensure fast block
+propagation to and from miners in mainland China.
+
+That is, instead of relying in random connectivity and being hampered by GFW issues, we propose
+to operate a small number of specialized nodes with the intent of enhancing network connectivity
+to our mining base.
+The specialized nodes, called boosters, will operate with extended meshes (at the gossipsub
+level) and utilize explicit peering agreements with large miners and exchanges.
+This will ensure that blocks (and messages) have fast network paths in and out of mainland China
+and improve the health of the filecoin network.
+
+#### Assumptions &amp; hypotheses
+
+- Network connectivity to miners in mainland China is a limiting factor.
+
+#### Impact
+
+- Enhance connectivity at the filecoin network.
+- Improve block and message propagation times in and out of China.
+- Eliminate late blocks and orphan mining.
+
+#### Internal leverage
+
+Reasonably High.
+
+#### Confidence
+
+Resonably Good.
+
+## Project definition
+#### Brief plan of attack
+
+In terms of code, there is a little work to be done:
+- make sure that the splitstore doesn't have any cache misses, so that boosters can operate
+  without a coldstore
+- create the pubsub profile for the booster role
+
+In terms of deployment, a small number (eg 4) of boosters should be sufficient.
+
+However, for the project to be successful we will have to reach out to miners and exchanges so as
+to establish direct peering agreements with our boosters.
+
+#### What does done look like?
+
+We operate a small number of boosters, explicitly peered with miners
+and exchanges, and have spiffy fast block/message propagation.
+
+####  What does success look like?
+
+- Block and message propagation times drop.
+- Late blocks and orphan mining is eliminated.
+
+#### Counterpoints &amp; pre-mortem
+
+Maybe the network connectivity to/from china is not the problem and this won't help much; but we don't
+think this is the case.
+
+#### Alternatives
+
+It's the best approach to enhancing network connectivity.
+The alternative would be for small miners and exchanges to talk to each others and establish their
+own explicit peering agreements and hope that a stable network emerges.
+However, we are much better positioned to create the necessary agreements and operate the boosters.
+
+#### Dependencies/prerequisites
+- splitstore; we need to run without a coldstore so that we don't have ticking bombs with network
+  storage in our boosters.
+
+#### Future opportunities
+
+Coupled with the pubsub obervatory we can gain better insight in the network.
+
+## Required resources
+
+#### Effort estimate
+- Development time should be small, 1 engineer for 1-2 weeks should complete the necessary code changes.
+- Infra would have to provision the booster nodes
+- Comms would have to reach out to exchanges and miners to establish the peering agreements.
+
+#### Roles / skills needed
+- Devlopment requires knowledge of splitstore and pubsub internals.