March 2024
·
6 Reads
Bandwidth is an essential resource for blockchain systems since BFT protocols require a series of communication steps to reach a consensus. Reducing available bandwidth will directly hurt system performance, especially in leader-based protocols where the leader carries the main communication overhead and becomes a bottleneck. Current approaches reduce the load on the leader by replacing its original star topology communication model with a tree. However, using trees increases the round latency and makes the consensus view change extremely complicated. In this paper, we propose Segmenta, a BFT consensus protocol leveraging slicing broadcasts to achieve low latency in the case of restricted bandwidth. We use the erasure code to slice the transaction blocks in the consensus proposals and provide verifiability with the Merkle tree. The origin blocks will be reconstructed with the participation of all replicas, thus reducing the leader’s network overhead. To avoid the increase in communication steps, the additional actions brought by the block shards broadcast are integrated into different consensus phases. Meanwhile, we propose a pipelined version of Segmenta, which further optimizes the transaction waiting time and lowers latency. We evaluate our protocol through a simple implementation in the testbed of Aliyun. Compared to Hotstuff, Segmenta shows a smoother and slower trend of latency increase and gets 4.6\(\times \) higher throughput while requiring only 17% of the latency in limited bandwidth (10 Mbps).
