LMPT: A Novel Authenticated Data Structure to Eliminate Storage Bottlenecks for High Performance Blockchains

Jemin Andrew Choi, Sidi Mohamed Beillahi, Srisht Fateh Singh, Panagiotis Michalopoulos, Peilun Li, Andreas G. Veneris, Fan Long

Abstract: We present the Layered Merkle Patricia Trie (LMPT), a performant storage data structure for processing transactions in high-throughput systems when compared to traditional Merkle Patricia Tries used in Ethereum clients. LMPTs keep smaller intermediary tries in memory to alleviate read and write amplification from high-latency disk storage. As an additional feat, they also allow for the I/O and transaction verifier threads to be scheduled in parallel and independently. LMPTs can ultimately reduce significant I/O traffic that happens on the critical path of transaction processing. Empirical results show that LMPTs can process up to $\times6$ more transactions per second on real-life ERC20 smart contract workloads when compared to existing Ethereum clients.