2.2 KiB
Resharding
In current design, the epoch is defined to be fixed length, the epoch length is a constant parameter BlocksPerEpoch. In future, it will be dynamically adjustable according to security parameter. During the epoch transition, suppose there are N shards, we sort the shards according to the size of active nodes (that had staking for next epoch). The first N/2 larger shards will be called active committees, and the last N/2 smaller shards will be called inactive committees. Don't be confused by the name, they are all normal shards with same function.
All the information about sharding will be stored in BeaconChain. A sharding state is defined as a map which maps each NodeID to the ShardID the node belongs to. Every node will have a unique NodeID and be mapped to one ShardID. At the beginning of a new epoch, the BeaconChain leader will propose a new block containing the new sharding state, the new sharding state is uniquely determined by the randomness generated by distributed randomness protocol. During the consensus process, all the validators will perform the same calculation and verify the proposed sharding state is valid. After consensus is reached, each node will write the new sharding state into the block. This block is called epoch block. In current code, it's the first block of each epoch in BeaconChain.
The main function of resharding is CalculcateNewShardState. It will take 3 inputs: newNodeList, oldShardState, randomSeed and output newShardState. The newNodeList will be retrieved from BeaconChain staking transaction during the previous epoch. The randomSeed and oldShardState is stored in previous epoch block. It should be noticed that the randomSeed generation currently is mocked. After the distributed randomness protocol(drand) is ready, the drand service will generate the random seed for resharding.
The resharding process is as follows: we first get newNodeList from staking transactions from previous epoch and assign the new nodes evenly into the N/2 active committees. Then, we kick out X% of nodes from each active committees and put these kicked out nodes into inactive committees evenly. The percentage X roughly equals to the percentage of new nodes into active committee in order to balance the committee size.