Blockchains are decentralized ledgers that replace trusted central authorities with verifiable distributed consensus. This decentralization has resulted in blockchains effectively becoming ‘slow and expensive computers’, but there are huge opportunities for architectural optimization across the entire blockchain software stack.

We begin this talk by outlining the scaling challenges from a systems researcher’s perspective, and discussing the bottlenecks faced in computation, storage, and network bandwidth. We then discuss how we optimized the blockchain storage layer using our novel verifiable database, the Quick Merkle Database (QMDB).

QMDB addresses longstanding state management bottlenecks with a streamlined design integrating Merkle tree storage and key-value stores. QMDB’s flash-friendly append-only design enables zero disk IO Merklelization with a minimal DRAM footprint. We demonstrate that QMDB scales linearly with IOPS on both consumer and enterprise hardware, reaching up to 2.3 million state updates per second. QMDB achieves throughput up to 6x higher than RocksDB, and up to 8x over NOMT, the previous state-of-the-art verifiable database.

The open-sourced code and preprint are available at https://layerzero.network/research/qmdb

QMDB is the first in a series of planned papers from LayerZero Labs that intend to challenge the assumption that high performance necessarily requires centralization.

Bio:
Daniel Wong is a Research Engineer in LayerZero Labs Research, a newly formed lab focused on scaling blockchain performance, and the youngest consortium member of the CMU Parallel Data Lab. Daniel completed his PhD thesis on ML for flash caching at the PDL in 2024. Like his advisor Greg Ganger, he has broad interests across systems research including caching, distributed storage, cloud computing, ML systems and security. He received his BA in Computer Science from the University of Cambridge. In his spare time, Daniel enjoys Singaporean food and making sushi.

More Info: https://pdl.cmu.edu/SDI/index.shtml