Ekiden

Conducted an experiment with 1000 clients, each sending 100 serialized requests to a compute node.

Batch compressed multiple state checkpoints into a single commit on the blockchain. Cached the latest state on compute nodes and used a writeahead log for state updates.

Because state is cached at compute nodes, compute nodes can opportunistically execute new transactions without waiting for a response from consensus nodes. Periodically, compute nodes asynchronously commit the state to the blockchain, as defined by the batch size. By separating contract execution from agreement on state, the layers can operate in parallel.

Compressed results from multiple requests into a single write to the blockchain - system has a total cost vastly less than that of on-chain execution.

A token transfer in 2ms and CryptoKitties breeding in 100ms

However, only one compute node, and no ability to do cross-contract calls.

With Tendermint as the consensus layer it achieved example performance of 600x more throughput and 400x less latency at 1000x less cost than the Ethereum mainnet.

Excels at dealing with computation intensive workloads i.e. Machine Learning