Characterizing Application Execution using the Open Community Runtime

Exascale and extreme-scale systems impose fundamental new requirements on software to target platforms with severe energy, data movement and resiliency constraints within and across nodes, and large degrees of homogeneous and heterogeneous parallelism and locality within a node. These challenges have led to the exploration of a diverse range of many-core processor architectures and memory hierarchies for future systems, that differ quite dramatically from current systems. As a result, there is still a lack of consensus as to what the main tenets should be for the execution model and low-level runtime system in future architectures. The Open Community Runtime (OCR) was created to engage the broader community of software and hardware researchers in identifying these underlying principles. While there is broad support for including dynamic task parallelism as one of the pillars of future execution models, there is currently little agreement on what else should be included. OCR proposes an approach to complete this picture by adding events and relocatable data-blocks as two additional pillars to build on, and shows how the three concepts (tasks, events, data blocks) can be combined in very general ways to support a wide range of higher-level programming constructs. In this paper, we focus on the use of OCR for application characterization. Despite the fact that the development of OCR is still at an early stage, we are fortunate that a large number of applications have already been implemented using the OCR APIs. We study the behavior of OCR implemenPermission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. RESPA ’15 Austin, Texas USA Copyright 2015 ACM X-XXXXX-XX-X/XX/XX ...$15.00. tations of these applications to gain insights into task durations, data block sizes, and access patterns. These insights can provide feedback to both application developers (on how to redesign algorithms for OCR-like execution models) and to hardware designers (to optimize for common characteristics of tasks, events and data blocks).