Technical Report: Dynamic X10: Resource-Aware Programming for Higher Efficiency

Static resource allocation, as common on supercomputers, decreases the overall system efficiency because most applications exhibit a varying degree of parallelism. Dynamic exclusive resource allocation can remedy this issue. For an X10 application, this means that the number of places can change dynamically at run-time. In this paper, we propose Dynamic X10, an X10 extension to support a changing number of places. We present an X10 framework for resource-aware programming, where resources are managed explicitly by the programmer. We show the necessary modifications to X10's runtime and standard library as well as an implementation of the proposed framework tailored to tiled many-core architectures. The required adaptions to existing X10 code are evaluated using applications from numerical mathematics.