ABSTRACT
State Machine Replication is a widely used technique to provide
fault tolerance and strong consistency. In this approach, all commands
are executed sequentially throughout the replicas. Aiming
to improve the system’s throughput, enhanced versions were proposed,
where independent commands can be executed in parallel.
One arising challenge, though, is how to balance the workload between
threads, while avoiding the need of synchronization between
them. This extended abstract proposes a technique to schedule requests
in a dynamic and efficient manner, using a workload graph to
keep track of access patterns and graph partitioning to decompose
and dispatch workload to the worker threads.