Thread by @EvanWeb3

On #SUI parallelism

Questions came up on this topic after my thread so here is a deeper dive.

Read on

1/x

https://twitter.com/EvanWeb3/status/1569414553322274818

Sui can process (sequencing + execution + commitment) multiple groups of txns in parallel. In this trivial example, (#1), (#2), and (#3, #4, #5).

2/x

The arch makes it possible that txns touching different shared objects can be sequenced in parallel via independent consensus instances.
This example - (#1), (#2), (#3, #4, #5) and (#6,#7,#8).

3/x

Some mistakenly believe only owned obj txns can be executed in parallel. That's wrong.

4/x

Sui can also execute the txns targeting the same shared object in parallel. This is not dissimilar to other parallel-execution blockchains, with a big difference.

5/x

After consensus, Sui can either use static scheduling (with object deps in the data model), or a dynamic (optimistic) approach, or a hybrid approach.

6/x

Once Sui validators go multi-worker, simple dynamic scheme doesn't work as txns and data need to be on the same machine. Sui can pull the possible conflict txns onto the same worker and then use optimistic concurrency locally on that worker. Many options to try

7/x

Having more static information about tx deps is strictly better than having less (particularly when such info is "free" rather than something you have to compute on the side)--having static info does not rule out using dynamic techniques on top! Sui data model FTW

8/x

TL;DR

1. Sui can leverage its data model to process multiple groups of txns in parallel
2. Sui can always execute txns targeting owned objs in parallel
3. Sui can execute txns targeting same shared objs in parallel with static and dynamic schemes

fin/