🤖 AI Summary
To address the high coordination overhead and the difficulty of pre-declaring dynamic read-write sets in cross-shard smart contract execution on DAG-based blockchains, this paper proposes the first lightweight, scalable execution framework tailored for DAG architectures. Methodologically, it introduces: (1) a dual-mode hybrid execution model (EOV/OE) unifying intra- and cross-shard transaction processing; (2) a runtime dependency-aware dynamic concurrency control mechanism eliminating the need for static read-write set declarations; (3) non-blocking shard reconfiguration coupled with a lightweight redistribution consensus protocol; and (4) a deterministic asynchronous DAG coordination protocol. Experimental evaluation on a 64-node deployment demonstrates a 50× throughput improvement over the serial Tusk framework, supports Turing-complete contracts with dynamic access patterns, and ensures continuous block production and consensus safety during reconfiguration.
📝 Abstract
Sharding has emerged as a critical technique for enhancing blockchain system scalability. However, existing sharding approaches face unique challenges when applied to Directed Acyclic Graph (DAG)-based protocols that integrate expressive smart contract processing. Current solutions predominantly rely on coordination mechanisms like 2PC and require transaction read/write sets to optimize parallel execution. These requirements introduce two fundamental limitations: 1) additional coordination phases incur latency overhead, and 2) pre-declaration of read/write sets proves impractical for Turing-complete smart contracts with dynamic access patterns. This paper presents Thunderbolt, a novel sharding architecture for both single-shard transactions (Single-shard TXs) and cross-shard transactions (Cross-shard TXs) and enables nonblocking reconfiguration to ensure system liveness. Our design introduces 4 key innovations: 1) each replica serves dual roles as a full-shard representative and transaction proposer, employing the Execution-Order-Validation (EOV) model for Single-shard TXs and Order-Execution (OE) model for Cross-shard TXs. 2) we develop a DAG-based coordination protocol that establishes deterministic ordering between two transaction types while preserving concurrent execution capabilities. 3) we implement a dynamic concurrency controller that schedules Single-shard TXs without requiring prior knowledge of read/write sets, enabling runtime dependency resolution. 4) Thunderbolt introduces a nonblocking shard reconfiguration mechanism to address censorship attacks by featuring frequent shard re-assignment without impeding the construction of DAG nor blocking consensus. Thunderbolt achieves a 50x throughput improvement with 64 replicas compared to serial execution in the Tusk framework.