One key area of research I'm doing with #cassandra is testing hybridization of deterministic & probabilistic consensus models.
Successfully doing so would provide an almost "best of both worlds" type model.
...
Successfully doing so would provide an almost "best of both worlds" type model.
...
Probabilistic (#nakamoto) has strong liveness guarantees but is never really safe.
Deterministic (classical) guarantees safety but is difficult to resolve both liveness and safety issues. Usually requiring some #godmode
Deterministic (classical) guarantees safety but is difficult to resolve both liveness and safety issues. Usually requiring some #godmode
The #cassandra approach has 2-phases.
The 1st phase is probabilistic promoting liveness to a strong guarantee.
The 2nd phase receives the output of the 1st phase and produces a classical 2f+1 safety threshold.
The 1st phase is probabilistic promoting liveness to a strong guarantee.
The 2nd phase receives the output of the 1st phase and produces a classical 2f+1 safety threshold.
The interesting piece here is the 1st phase. If say a portion of the network suddenly went offline, there was be a starvation of vote power.
The 1st phase can continue to make progress because it's not reliant on the 2f+1 threshold.
...
The 1st phase can continue to make progress because it's not reliant on the 2f+1 threshold.
...
Without a 2f+1 vote power threshold, the 2nd phase can not use the output of the 1st to commit and provide a guarantee on safety.
There is a mechanism in the 1st phase to increment the vote power of active validators on each round of progress.
...
There is a mechanism in the 1st phase to increment the vote power of active validators on each round of progress.
...
Eventually the increasing vote power of the active validators "dilutes" away the influence of the departed ones and viola!
The output of the 1st phase can now produce a 2f+1 quorum the 2nd phase need to do its thing & reverting from probabilistic safety to deterministic again.
The output of the 1st phase can now produce a 2f+1 quorum the 2nd phase need to do its thing & reverting from probabilistic safety to deterministic again.
This morning I implemented the final module needed to test the theory.
I ran the network for a while & killed 90% of it.
The remaining built a large quantity of probabilistically safe transactions, until suddenly the 2nd phase woke up & processed the lot!
#winning $exrd
I ran the network for a while & killed 90% of it.
The remaining built a large quantity of probabilistically safe transactions, until suddenly the 2nd phase woke up & processed the lot!
#winning $exrd
As far as I'm aware, this kind of hybridization has not been done before successfully.
#cassandra will be not only the first, but will be doing so without breaking the #trilemma or #quadrilemma
#cassandra will be not only the first, but will be doing so without breaking the #trilemma or #quadrilemma
• • •
Missing some Tweet in this thread? You can try to
force a refresh
