GridSMR: Causal Compression for Sharded Blockchains

GridSMR is a sharded blockchain architecture that achieves horizontal execution scalability while enabling dependent cross-shard operations to progress within a single block through Causal Compression. By employing an execute-before-agree model, GridSMR localizes atomicity to individual accounts and executes cross-account work asynchronously, allowing consensus to validate and commit the resulting schedule only after execution.

Key performance metrics include scaling single-validator execution to 1.07M requests/s and four-validator execution to 193K committed requests/s. This architecture reduces the latency of 16-hop causal chains by 8.2× compared to deferred execution models, significantly outperforming existing systems like NEAR testnet where each causal hop typically requires an additional agreement round.

GridSMR is presented as a sharded blockchain architecture aimed at preserving the horizontal scalability of sharding while reducing the latency penalty caused by dependent cross-shard operations. In many existing sharded systems, a transaction or contract step that depends on the result of a previous step in another shard requires explicit coordination, often causing the operation to propagate through multiple blocks and making end-to-end latency grow with causal depth. GridSMR addresses this by localizing atomicity to individual accounts rather than requiring broader cross-shard coordination for every dependent transition, and by using “causal compression” to allow a chain of dependent cross-shard operations to progress within a single block.

The key insight is that much of the coordination cost in sharded blockchains comes from treating cross-shard causality as a repeated, block-by-block synchronization problem. By compressing causal dependencies and executing dependent steps together, GridSMR lets shards continue to scale execution in parallel while still maintaining the ordering and state-consistency requirements needed for dependent operations. This design suggests a shift from global or heavy-weight cross-shard atomicity toward finer-grained, account-local atomicity combined with a more efficient representation of dependencies, potentially enabling complex multi-step workflows without sacrificing throughput.

This matters because sharding is one of the main paths to scaling blockchain execution, but cross-shard composability has often been a practical bottleneck. If GridSMR can reduce the latency of dependent operations from being proportional to causal depth, it would make sharded chains more usable for applications that require multi-step cross-shard interactions, such as smart contracts, decentralized finance flows, payments, and multi-party coordination. More broadly, the work is significant for the design of scalable distributed ledgers: it proposes a way to keep the parallelism benefits of sharding while making causal dependencies less expensive, which is a central challenge in moving blockchains from single-chain execution models to high-throughput, multi-shard systems.