StreamChain

Abstract

Processing at block granularity and blockchains seem inseparable. The original role of blocks is to amortize the cost of cryptography (e.g., solving proof-of-work) and to make data transfers more efficient in a geo-distributed setting. While blocks are a simple and powerful tool for amortizing these costs, today in permissioned distributed ledgers, that are often neither geo-distributed, nor require proof-of-work, the benefits of operating on blocks are overshadowed by the large latencies they introduce. Our proposal is to switch the distributed ledger processing paradigm from block processing to stream transaction processing and rely on batching (i.e., block formation) only for amortizing the cost of disk accesses for commit operations. This paradigm shift enables shaving off end-to-end latencies by more than an order of magnitude and opens up new use-cases for permissioned ledgers. We demonstrate a proof-of-concept of our idea using Hyperledger Fabric, achieving end-to-end latencies of less than 10ms while maintaining relatively high throughput, namely close to 1500 tps.