A journey through the underbelly of eth2’s final call before takeoff.
By Elias Simos and Sid Shekhar.
About this work: Elias is a Protocol Specialist at Bison Trails. Sid leads Blockchain Research at Coinbase. This report was an independent collaborative research effort to review the overall performance of the Medalla Testnet, answering the EF’s call for submissions on the eth2 Medalla data challenge.
All data, scripts, and notebooks that powered this analysis are available on github. The majority of the data we used came from the Prysm API via beaconcha.in.Table of Contents
Highlights We were able to group together ~85% of all validators in Medalla. Over 2/3 of all the validators on the network are concentrated between at most 22 groups, many of which we were able to identify as developers, infrastructure providers and Staking as a Service firms. While we weren’t able to directly identify exchanges or custodians, we singled out 4 entities that could fit the profile. We were further able to identify the client choice of over 1/3 of the network’s validators. As expected, Prysm takes the lions share, with Lighthouse improving the most over the 14500 epochs we investigated, and looking ready to contend. We provide a definition for “normal” vs “abnormal” network conditions. Under “normal” conditions, 3/4 of block slots available were proposed and 1/4 was missed. The distribution of proposed blocks per client matched the overall distrubution of clients, while in the distribution of missed slots per client, Nimbus features a lot more prominently. Client syncing issues have been the most probable root cause for over 90% of attester and proposer slashings in Medalla. All attester slashings were attributed to double votes, most of which shared the same source and target epoch. We observed no penalties enforced for surrounding voting. We introduce aggregate inclusion delay as a novel ...