Ethereum Foundation

The Ethereum Foundation (EF) is an organization that supports and works towards the long‑term success of the Ethereum network . It maintains key components of the network and also funds research, infrastructure, education, tooling, and community programs to advance Ethereum’s scalability, security, and adoption.

My work

I was on the Peer to Peer (P2P) Networking Team. P2P networking is the networking abstraction that supports communication between nodes in the network using libp2p. The main theme of my work was improving observability of the network layer of Ethereum. This came with three major workflows:

1) Simulation and Benchmarking

Motivation

The networking layer of Ethereum, libp2p have various implementations of in different languages by different groups. Testing within a single implementation can be done relatively throughly but the challenge is testing between different implementations. There is also the challenge of Ethereum nodes not being managed by a central entity. Therefore, it is hard to test certain behaviors and scenarios that may occur simply because you don’t control all the nodes.

Solution

The foundation has a crawler that crawls nodes on mainnet and exports data about a node such as which client/implementation it is running, the IP address, geo location, autnomous system number, etc. Using this list of nodes in the network and other data sources such as latencies between autnomous systems/geo locations and bandwidth limitations of a node, it is possible to replicate the network graph of mainnet locally. This is powerful because you can create certain scenarios that you wouldn’t be able to on the actual network and test the interoperability of differnet implementations of libp2p throughly. Luckily, there is already a tool for simulating called shadow. shadow, from its documentation, “is a discrete-event network simulator that directly executes real application code, enabling you to simulate distributed systems with thousands of network-connected processes in realistic and scalable private network experiments using your laptop, desktop, or server running Linux.”

My work

Source up to date and reliable inter-autnomous system latency measurements. RIPE Atlas has streaming API that continously streams ping measurements between probes across the world. I built a program that subcribes to this stream and exports the data to a data store. https://github.com/wbjin/PingStreamer
Create a network graph using the list of source nodes and latency measurements that can be fed into shadow. Because there are limitations on how many processes you can realistically run for a shadow simulation, I had to come up with a way to as closely replicate mainnet’s network graph without blowing a computer up

2) Telemetry and Analytics of libp2p

Motivation

The Ethereum Foundation and many of the clients already had a robust observability framework that exported telemetry data from client nodes to a central data store. However, there wasn’t as much analytics done on the telemetry data. Basically, we weren’t really looking into the data.

Solution

Make cool graphs

My work

Analyze the robust telemetry data to identify any patterns in the network. Things we looked for were

Any irregularities
Average and tail latencies
Network graph of mainnet
Health of the network

Here are some visualizations

3) Expanding telemetry collection of the networking layer

Motivation

All of the above is powered by robust telemetry data. While the current telemetry pipeline has a lot insightful data, it doesn’t capture all of the available metrics from libp2p implementations. Getting even more telemetry will allow more robust and a more complete view of the network.

Solution

There are three different ways telemetry data is collected: crawlers, nodes that sit in the network and act as a regular node but also export telemetry data, full clients that export telemetry data.

My work

Create new gRPC service for collecting node resource usage such. Key metrics include CPU usage, memory usage, jitter.
Expanding existing gRPC services and writing new kafka translations to collect more libp2p metrics.