Csep-Reading-8B

Hoplite: efficient and fault-tolerant collective communication for task-based distributed systems

This paper does a great job of explaining exactly how Hoplite improves on task-based distributed systems frameworks. To be honest though, this paper feels very tangential compared to other papers we've read. It deals with an application that runs over a network, but doesn't really detail how the network is involved, other than the small detail in Section 4: Implementation that explains data is transfered via gRCP (which runs on HTTP 2) and, for transfer requests like Get, a direct TCP connection. We don't get any explanation for these choices.