owned this note   changed 6 years ago  
Linked with GitHub

Ethereum 222.0 Workshop Notes

All workshop attendees are encouraged to contribute their own notes to this document. This is a shared space for us to help further the spec. This document format is admittedly abysmal, so feel free to restructure. It's the best option at the moment for real time notetaking.

Ameen Soleimani (CEO SpankChain)'s Talk Notes (State of Eth2 Implementation)

  • There is more that goes in to building a client than just implementing the spec. Architecture, design, optimization, and other low-level implementation decisions must be made, so the spec does not truly represent the body of work necessary to have clients taking to each other on an Eth2 test net.
  • Before we move towards a testnet, we need consistent cross client tests.
  • Spec updates should follow a versioning paradigm, so there is a consistent reference for implementers to know what is expected of clients at a given time.
  • While the spec has always been open and available, EF did not explicitly consult potential Eth2 implementers when formulating the mainnet rollout plan. EF Researchers followed best practices and in general believed the Eth2's phased approach to mainnet rollout captured the opinions of development teams.
  • Implementation may have started too early, and the full complexity of the spec, even the Phase 0 Beacon Chain spec, was not fully understood.
  • What are the pros and cons of having a lead or other "centralized" coordination role, if it is being done with the consent of the implementers?
    • Cons: There is a non-zero possibility of a central planner overstepping their authority and taking the spec in an unpopular direction.
    • Pros: If everyone works on their own pet projects or favorite portion of the spec, it's difficult to form a cohesive development community.
  • Some central planning was necessary in order to kick off the research process. As more people became familiar with and contribute in the direction, it is easier to open up to more decentralized planning.
  • Why did client developers start so soon? Did they feel as though they had to implement Eth2 immediately? Perhaps there were some miscues and signaling problems to begin with.
  • There is now an apetite in the community for people to contribute and help Eth2 more faster and better.
  • The sheer number of teams working on Eth2 at the moment introduces significant coordination overhead.
  • Eth2 is underestimating competitive threats. There are many other blockchains teams whose success depends on convincing the next generation to build their ideas on a platform not Ethereum.
  • Eth2 roadmap is technically ambituous but still moving too slowly.
  • It shouldn't be expected/necessary that all teams launch at the same time.

Mikerah (ChainSafe)'s Talk Notes (Why should we care about validator privacy?)

  • GitHub Stuff
  • Is there a rational for using Kademlia? Answer: It's already implemented in libp2p.
  • In structured networks long lived identities are necessary feature for implementation. Can we consider unstructured networks?
  • Ideal is to not tie node identity to validator identity in the protocol. Not tightly coupling this allows the validator more freedom in implementation.
  • Wire protocol needs to be well specified so the clients actually know how to communicate.
  • Gossip can be inefficient because the same transaction data is sent even to clients that have already received it.
  • What are the drawbacks to using the protocols that are already around today? i.e. devp2p Answer: It's poorly specified.
  • After significant research and benchmarking and the libp2p team offering closeness with the EF, libp2p was chosen.
  • What are alternatives to Kademlia? Answer: Bitcoin uses a unstructured network which is mostly hueristic based.
  • Minimal propagation time of new messages should be preferred in blockchain networks over bounded lookup times.
  • Is using an unstructured network even possible if your goal is to index nodes?
  • Validators are reassigned at each epoch, so a structured network will have to recalcuated at a per-epoch basis, correct? There is per-epoch shuffling, but also the notion of persistent committees in Eth2.
  • Will contracts on different shards be able to speak to eachother? Answer: Research is underway on a async crossshard communication protocol built in to base layer as well as a number of synchronous layer 2 protocols being investigated.
  • Papers that were mentioned

Vanessa Bridge (PegaSys R&D | ConsenSys)'s Talk Notes

  • Wittgenstein Repo on GitHub
  • A simulation framework for testing some PoS and consensus algorithms. It is still in development, but hopefully others find this useful for tuning Eth2 network parameters and making sensible design decisions.
  • Used to simulate Casper IMD.
  • As of current protocols must be written in Java to make use of Wittgenstein simulation.
  • There are no plans for JNI bindings to faciliate testing protocols build in other languages which expose shared objects as of yet, but the implementation team is open to the idea given demand.

TestNet & Wire Protocol Discussion Notes

Session Takeaways
What does it take to achieve interop?

  1. We need to define these interop tasks:

    • Discovery (see py-libp2p team's summary below)
      • Felix from the EF is working on the Discovery Protocol
    • Connection Handshaking
      • Encryption?
      • Authenitication?
      • Compression?
      • Peer reputation?
    • Application Protocol
      • How will we propogate blocks and attestations?

  2. Now that there is an initial spec release, the major priority is creating:

    • test vector formats
    • test vector generators
      well designed tests will help us achieve interop sooner.

    Danny's TODO:

    • to take a stab at designing the wire protocol in the next couple of days.
    • is going to add some notes to the spec on Network Topology

py-libp2p Peer Discovery Discussion Notes

Slide Deck
Session takeaways

  • Discovery protocol should return ENRs (Ethereum Node Records) or other forms of node metadata, instead of just networking information.
    • Easier to discover node capabilities, e.g. shards associated with a node
  • Endpoint verification needs more thought
    • Making the "bonding" process described in presentation more resilient to network faults
  • Non-Kademlia options should be explored.

Additional info:

Select a repo