Data Dissemination for DA Scaling

[l-monninger]

Summary

A data dissemination layer can be used to scale throughput against the DA.

DAs, Data Dissemination, and Throughput

Total ordering DA throughput is effectively limited by the layer's byte per second sequencing capabilities. However, applications built on a DA will typically best measure performance in terms of transactions per second or gas per second. It is, in fact, the representation of data which results in these units which best correlates with actual throughput. One way to increase throughput on said applications is, thus, to shrink the representations which correlate with said units.

In a proposer-based blockchain--as Movement Suzuka is structured to be from the standpoint of the DA--the block can be identified as the fundamental unit which need be sequenced. Its representation can be minimized in a cryptographically safe manner with a cryptographic hash. The contents of said can then be recovered over a peer-to-peer network, i.e., via data dissemination.

In Suzuka's use of Celestia--which has roughly 6 mb/s throughput--this approach would allow us to scale to throughput against Celestia far exceeding the 170,000 TPS capabilities of Aptos VM. In fact, when ignoring the time spent constructing and digesting blocks prior to Celestia, throughput is unbounded as blocks become arbitrarily large and still remain the same 256 bit representation or similar.

This approach is not novel. In fact, Aptos has developed Quorum Store, to do more less exactly this: https://medium.com/aptoslabs/quorum-store-how-consensus-horizontally-scales-on-the-aptos-blockchain-988866f6d5b0

Shifting Bottlenecks

The data dissemination approach shifts the problem of throughput primarily to the following operations:

1. Gossip and verification of digested data.
2. Data digestion (hashing).
3. Proposals.

Minimizing Trust Assumptions

It must be assumed that data shared during dissemination may be untrustworthy.

1. That the data provided via dissemination matches the data digested is verifiable via checking a computed hash matches the disseminated cryptographic hash.
2. To ensure provenance and the capability for penalizing maliciously constructed data, data disseminated should be cryptographically signed.