The documents distributed by this server have been provided by the contributing authors as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.
Publications of SPCL
|Efficient networking and programming of large-scale computing systems|
(Presentation - presented in Palo Alto, CA, USA, Jun. 2015, )
AbstractWe will discuss efficient techniques for large scale datacenter networking. We start by introducing a high-performance cost-effective network topology called Slim Fly that approaches the theoretically optimal network diameter. Slim Fly is based on graphs that approximate the solution to the degree-diameter problem. We analyze Slim Fly and compare it to both traditional and state-of-the-art networks. Our analysis shows that Slim Fly has significant advantages over other topologies in latency, bandwidth, resiliency, cost, and power consumption. Finally, we propose deadlock-free routing schemes and physical layouts for large computing centers as well as a detailed cost and power model. We continue our discussion by considering the endpoint interface. Here, we propose remote memory access programming which offers abstractions to coordinate directly accessible distributed memory domains. We continue by showing how RMA programming simplifies the design and tuning and introduce MPI-3's RMA semantics as a particuler example. We discuss our reference implementation for Cray machines foMPI and demonstrate results with up to half a million processes. We conclude the talk by addressing producer-consumer synchronizations in task-based runtime environments and the new proposal of notified access. Overall, we advocate RMA as a potential programming model for scalable systems ranging from single-die multicores to large-scale supercomputers.