Sharded blockchain simulator hits 1.6M TPS at 256 shards
New SimPy model reveals limits of sharding as coordination traffic dominates gains.
A new simulation framework from Om Amit Gandhi and Ioan Raicu tackles the scalability limits of public blockchains by modeling sharded architectures. Built on SimPy, the open-source discrete-event library, this configurable simulator accounts for mining, verification, inter-shard coordination, block dissemination, and communication overhead. It reveals that simply increasing block size or decreasing intervals leads to higher propagation delays, forks, and stale blocks—sharding provides a better path forward.
Key results show 1.6 million transactions per second (TPS) at 256 shards in a local datacenter-like network, and about 0.6M TPS under a global WAN topology. However, the gains are not unlimited: beyond a certain number of shards, coordination traffic and synchronization overhead begin to dominate, leading to diminishing returns. This finding is critical for designing next-generation blockchain systems, suggesting an optimal shard count exists where throughput gains outweigh overhead. The 4-page technical report, to be presented at GCASR 2026, provides a practical tool for researchers and engineers evaluating trade-offs in distributed ledger performance.
- SimPy-based simulator models mining, verification, inter-shard coordination, and block dissemination
- Achieved 1.6M TPS at 256 shards in local datacenter, 0.6M TPS in global WAN
- Diminishing returns observed after optimal shard count due to coordination and network overhead
Why It Matters
Provides a practical benchmark for understanding when sharding helps—and when it hurts—real-world blockchain throughput.