We’ve spent the last nine months testing the life out of this thing. The year began with huge, long-running single client testnets: Sapphire, Topaz and Onyx networks run by Prysmatic Labs. In April, there were small multi-client networks: Schlesi, Witti and Altona – all named after subway stations, in keeping with Ethereum testnet tradition.
And then the big one, the Medalla testnet. Named after Medalla Milagrosa on the Buenos Aires Underground, it has been running for over two months, with four different client implementations involved throughout that time. It continues to run today with over 50,000 validators actively participating, making it one of the largest decentralized consensus networks in existence.
Progress has not all been smooth. A few days after the start of the Medalla testnet, one of the clients suffered that disrupted the chain for a few days. But this is what testnets are for. We kept the chain running and were able to bring it back to full health, with a slew of lessons learned.
Among them, client diversity is important. If we want the beacon chain to be resilient, no single client implementation can dominate. As Danny Ryan, a core researcher at the Ethereum Foundation, , “The incident on Medalla was significantly amplified by the failure of the dominant Prysm client, and as we move toward mainnet, we, as a community, must consciously seek to remedy this.”
Four high-quality, audited and battle-tested clients are currently available to run at beacon chain launch: Teku, Lighthouse, Nimbus and Prysm. Each has its own flavor and target user base. For example, , the Eth 2.0 client from ConsenSys, has been designed and built primarily with institutional and professional stakers in mind (although I shall be running it at home), with extra security tools such as a remote signing service and a prevention service.
Client teams also learned to agree on common standards for migrating information between their implementations. This allows stakers to safely switch quickly between clients and will greatly help with incident recovery in future.