2016 Poster Sessions : Integrating External Resource into a Task-Based Model

Student Name : Zhihao Jia
Advisor : Alex Aiken
Research Areas: Computer Systems
Accessing external resources (e.g., loading input data, checkpointing snapshots, out-of-core proccessing) can have a significant impact on the performance of supercomputer applications. However, no existing programming systems for high-performance computing directly manage and optimize these external accesses with respect to computation. As a result, users must explicitly manage external accesses alongside their computation at the application level, which can result in both correctness and performance issues.

We address this limitation by introducing Via, a task-based programming model with semantics for external resources that directly allows the application to describe its access requirements to external resources with respect to its computation. Via incorporates external I/O into a deferred execution model, reschedules external I/O to overlap I/O with computation, and reduces external I/O when possible. We evaluate Via on three microbenchmarks representative of important workloads in HPC and a full combustion simulation package, S3D. We demonstrate that the Via implementation of S3D imposes only 2-3% overhead for external I/O, compared to 15-76% and 2-12% achieved by the Legion and the Fortran implementations, respectively.

Zhihao Jia is a third year PhD student working with Alex Aiken. He is actively contributing to the Legion programming system, which is a data-centric parallel programming system for writing portable high performance programs targeted at distributed heterogeneous architectures.