Great work @protolambda and others!
This EIP suggests “a target of ~1 MB per block and a limit of ~2 MB”. I believe StarkWare’s First “Big Red Dot” and Second “Big Red Dot” mainnet experiments and analysis from July 2019 and Jan 2020, respectively. This analysis showed that when creating a series of “large” mainnet blocks (45KB-1.46MB) throughout a single day, that it had no appreciable affect on uncle rates.
This is an important and valuable analysis, but I think that there are subsequent simulations and analysis warranted before moving into the realm of regular 1MB blocks on mainnet.
A few things to consider
- Does the shift from PoW to PoS have any impact on the prior analysis?
- In PoW, there are likely in the small number of dozens of consensus forming nodes (miners) rather than on the order of thousands in the current beacon chain network. Orphaning in PoW is thus the question of whether these large blocks make it to the other (small number of) miners quickly rather than to all user nodes or to all stakers in PoS
- 12s slot times in PoS might actually help prevent orphaning compared to PoW (a block is not likely to be orphaned if it makes in sub 12s times whereas the threshold is likely lower in PoW). But there is another important time, 4s – the attestation time into the slot. If large blocks regularly make it to the network after 4s, we’d see “incorrect head votes” for many attestations even though the block still makes it into the chain. This would be an indicator of non-optimality in the consensus and some loss of revenue for validators. But would potentially result in orphaning in future block-slot PBS fork choice rules which have a stricter reaction to these missed head attestations.
- Does the shift from devp2p block gossip to libp2p block gossip have any impact on the expectations of load, propagation time, etc
- Does the uncle rate actually capture the quality of service required here?
- Even though blocks were not orphaned in the Big Red Dot experiments, are user nodes receiving and processing these blocks in a timely manner? or do the small set of mining nodes have some asymmetric bandwidth available and/or privileged position in the network (highly connected, multiple sentry nodes, etc)
- What does the minimum network speed/bandwidth become in a 1MB block regime? Does this push out some class of user or locality? e.g. at 10mbps, 1MB block transfers take ~1s per hop, whereas a 100mbps, they take 0.1s
- What becomes the minimum monthly data-cap required for such regular blocks (taking into account data-tx mempool requirements and the gossip amplification factor)?
I don’t intend to nay-say the 1MB suggestion, but I want to highlight there are some pen-and-paper calculations and likely some network simulations in order beyond the prior Big Red Dot analyses to tune this number.