Snowflake to Avalanche: A Novel Metastable Consensus Protocol Family for Cryptocurrencies. This was the title of a whitepaper that was delivered anonymously to the folks at InterPlanetary File System (IPFS) earlier this week.
The paper, which was also posted on some online forums, has already caused quite a stir.
"Someone dropped this paper on IPFS and some IRC channels yesterday," wrote Emin Gun Sirer, Cornell professor and co-author of A Call for a Temporary Moratorium on The DAO. "It describes a new family of consensus protocols that combines the best of Nakamoto consensus with the best of classical consensus. Huge breakthrough."
The authors, known only as "Team Rocket," explain how the new protocol family can be used to "construct the core of an internet-scale electronic payment system," which purportedly has been tested under experimental conditions to a throughput level of 1300 transactions per second, with four-second confirmation latency.
Moreover, the authors note the novel implementations "scale well compared to existing systems that deliver similar functionality."
The consensus algorithms, Snowflake, Snowball, and Avalanche, offer a new approach to reaching consensus. More traditional consensus protocols rely on all-to-all communication to assure every node in a system is reaching the same decisions. But according to Team Rocket, these methods "are quite costly, wasteful, and limited in performance."
Taking inspiration from "gossip algorithms" (similar to those used in Hashgraph), the algorithms proposed by Team Rocket (which are not asynchronous Byzantine fault-tolerant) gain their security features "through a deliberately metastable mechanism."
Operating by repeatedly sampling the network at random, Team Rocket's system steers correct nodes toward the same outcomes.
Like Nakamoto consensus, the new protocol family offers a probabilistic safety guarantee. However, unlike Nakamoto consensus, Team Rocket alleges its protocols are "green, quiescent and efficient." They don't rely on proof-of-work and don't consume electricity when decisions are not being processed.
The proposed payment system, which has been implemented in roughly 5000 lines of C++, might be an entry point for the ecosystem to discuss tradeoffs between security and functionality or speed.
Team Rocket acknowledges its security solutions work at the cost of not being able to guarantee liveness for conflicting transactions.
Regardless, with so much of the distributed ledger revolution currently bottlenecked by scalability challenges, a Bitcoin-like payment system that achieves 1300 TPS might be food for thought if nothing more.