Computation Handling

Computations are defined and scheduled inside of MXEs. Clusters execute computations for their associated MXEs. Since the Arcium Network itself is stateless, the execution of all computations – even those dedicated to a single MXE – can happen fully parallelized. A Cluster can execute any number of computations concurrently – the only bottleneck being the hardware capacity of the least capable node in a Cluster. Network-wide parallelization is further boosted by the Network not requiring to perform work over shared state (not even neccessarily for individual MXEs), the Arcium Network can instead be viewed as a highly parallelized blackbox, performing computations with arbitrary runtime and complexity in parallel. While MXEs can be configured to use multiple Clusters, each scheduled execution of a computation occurs within a single one of the Clusters associated with the MXE. MXEs that use multiple Clusters, to improve their availability guarantees by not entirely relying on the responsiveness of any one of the individual Clusters they use, can respond to Cluster unavailability faster (e.g. by immediately retrying a failed computation using a different Cluster already associated with the MXE).