Improving Fault Tolerance in Blockchain Sharding using One-to-Many Block-to-Shard Mapping

The goal of sharding in a contemporary Blockchain system is to increase throughput linearly in proportion to the number of shards. This is achieved in practice by one-to-one mapping of each transaction block to a shard, with the assumption that each shard is 'perfect' and hence cannot fail. The notion of perfection is achieved by forming shards that have negligible failure probabilities. This contemporary approach to blockchain sharding has two drawbacks: (1) shards tend to be large in size to maintain low failure probability, which can negatively affect performance and throughput; (2) the 'perfect' shard assumption can easily be breached if any shard becomes faulty, which can fail an entire blockchain system because there is no fault-detection mechanism during transaction-processing (i.e., faulty blocks approved by faulty shards may only be detected after being appended to the blockchain). To overcome these drawbacks, this paper presents a multi-round consensus scheme which adopts one-to-many mapping of a transaction block to k shards, followed by a second consensus round among the k shard leaders (inter-shard consensus) in an epoch to validate and commit a transaction block with finality. In return, the following are achieved: (1) possibility of increased fault tolerance, despite using smaller shard sizes, because the collective failure probability of a group of k small shards can be much lower than the failure probability of an individual larger shard with the proper selection of the values of k and other parameters; (2) capability of faulty block detection with high probability during transaction processing; and (3) relaxation of the 'perfect' shard assumption so that the system can be tolerant to more faulty shards and still maintain safety. Detailed theoretical analyses are presented which demonstrate the benefits of such a multi-round block validation approach over contemporary approaches in terms of achieving better fault tolerance without compromising on throughput.