Review of the PARSEC paper

Hmm… I don’t know. From my understanding the algorithm should run without forking detection. But on the other hand forking detection is a kind of byzantine fault detector, which is good to have for punishment.

3 Likes

Hi Andre,

Thanks for your input.

Now what happens if Alice creates a fork with a new event, that references to one of these old event, where order was already reached over? Doesn’t it mean, that all work from that old event on has to be recomputed?

I think you’re describing this as a spamming action from Alice?

A few things to note: if an event has both sides of the forks in its ancestry (which should probably be the case if she’s forking from a long settled event), it sees no event by Alice at all. This allows PARSEC on its own to maintain all its proofs. It may also help reduce the impact of this spam (not 100% sure on this).

Also, as soon as the fork is detected, consensus on kicking Alice out will be initiated and she will be kicked out as soon as consensus is reached, so the cost of this kind of spam is high as Alice first had to spend enough resources to become an elder so she could even participate in the consensus decision (SAFE Network specific, not PARSEC specific).

Currently, we are kind of generously recomputing past meta-data, though. Caching and pruning it is something we will spend more energy on later in this milestone; when looking at performance optimizations.

16 Likes

Hi @dirvine, oh is there another forum, who can join? Sounds like it could be a great read.

1 Like
2 Likes