SISSLE in consensus-based Ripple: Some Improvements in Speed, Security and Last Mile Connectivity
This work addresses security and performance issues in Ripple for banks and payment providers, offering incremental enhancements to an existing system.
The paper tackles improvements to the Ripple cryptocurrency's consensus algorithm, achieving significant gains such as a ≥99.67% reduction in double spend attacks, a 1.71x increase in fault tolerance to ≥34.21% malicious nodes, and speedups of up to 98.22x in information propagation.
Cryptocurrencies are rapidly finding application in areas such as Real Time Gross Settlements and Payments. Ripple is a cryptocurrency that has gained prominence with banks and payment providers. It solves the Byzantine General's Problem with its Ripple Protocol Consensus Algorithm (RPCA), where each server maintains a list of servers, called the Unique Node List (UNL), that represents the network for that server and will not collectively defraud it. The server believes that the network has come to a consensus when servers on the UNL come to a consensus on a transaction. In this paper we improve Ripple to achieve better speed, security and last mile connectivity. We implement guidelines for resilience, robustness, improved security, and efficient information propagation (IP). We enhance the system to ensure that each server receives information from across the whole network rather than just from the UNL members. We introduce the paradigm of UNL overlap as a function of IP and the trust a server assigns to its own UNL. Our design makes it possible to identify and mitigate some malicious behaviours including attempts to fraudulently Double Spend or stall the system. We provide experimental evidence of the benefits of our approach over the current Ripple scheme. We observe $\geq 99.67\%$ reduction in opportunities for double spend attacks and censorship, $1.71x$ increase in fault tolerance to $\geq 34.21\%$ malicious nodes, $\geq 4.97x$ and $98.22x$ speedup and success rate for IP respectively, and $\geq 3.16x$ and $51.70x$ speedup and success rate in consensus respectively.