HanDist

In this article, we present our relocatable distributed collection library. Building on top of the AGPAS for Java library, we provide a number of useful intranode parallel patterns as well as the features necessary to support the distributed nature of the computation through clearly identified methods. In particular, the transfer of distributed collections’ entries between processes is supported via an integrated relocation system. This enables dynamic load-balancing capabilities, making it possible for programs to adapt to uneven or evolving cluster performance. The system we developed makes it possible to dynamically control the distribution and the data flow of distributed programs through high-level abstractions. Programmers using our library can, therefore, write complex distributed programs combining computation and communication phases through a consistent API. We evaluate the performance of our library against two programs taken from well-known Java benchmark suites, demonstrating superior programmability and obtaining better performance on one benchmark and reasonable overhead on the second. Finally, we demonstrate the ease and benefits of load balancing and a more complex application, which uses the various features of our library extensively.

In this paper and we introduce the global load balancer integrated into our distributed collections library for the APGAS for Java programming model. Inspired by the lifeline-based Global Load Balancer scheme and our load balancer makes it possible for programmers to perform actions on the objects recorded into a distributed collection while allowing the library to relocate entries between hosts if load imbalances occur. The programming model we adopt introduces minimal impact on the legibility of programs and with the regions of the program where entries of a distributed collection may be relocated clearly identified. Internally and we implement a hybrid scheme which balances the load both between the threads on a host and between the hosts participating in the computation. This allows for multiple concurrent computations on multiple collections with individual termination detection. We evaluate the performance of our integrated Global Load Balancer and its ability to handle various situations on a many-core supercomputer.

The increase in the number of cores available in modern processors makes it important for implementations to maximize their use within a node by overlapping communication and computation. However, when the dependencies between communication and computation are complex and evolve over the course of the execution, their implementation becomes tedious and may lead to bugs. In this paper, we focus on spatial simulation where elements move over time due to interactions with other surrounding elements. We propose a Distributed Cell Set Library that manages the distribution of the simulation space using unit areas (cell units). We make it possible to manage the granularity with which cell communication may overlap with computation, and for multithreaded computation. In addition, we make it possible to describe the relationships between inter-node communication and the pending computation in cell units. For evaluation, we introduce the implementation of a two-dimensional molecular dynamics simulation using our library. We show that using our computation and communication overlapping method, the delays to reach global synchronization that are necessary in the original implementation can be avoided.