Efficient Message Passing Support for Asynchronous Many-Task Systems

The communication behaviors of asynchronous many-task systems are usually irregular and happen in a multithreaded environment. This leads to some mismatches between what task systems want and what their underlying communication libraries offer, leading to inefficiencies such as poor multithreaded performance, unnecessary memory copies and messages, unpredictable background task processing, and inefficient polling for completion. We are developing a low-level communication library, Lightweight Communication Interface (LCI), to explore ways to eliminate these mismatches and provide direct communication support to task systems. LCI’s features include (a) flexible communication primitives and signaling mechanisms (b) better-multithreaded performance (c) explicit user control of communication resources. We are working on creating an LCI parcelport for HPX. We will report our current progress in this poster.