Reducing power consumption through dynamic load balancing

Click HERE to download presentation

Abstract : Although parallel application developers do their best in distributing the data or instructions in the different processes, load-imbalance is everywhere due to reasons not controllable, and what is more, is every time  worse with the current trend in HPC systems where more and more cores are available.

Load can be redistributed at given simulation steps to reduce this issue, but still we can observe load unbalance at small granularity which is difficult to reduce from the application. At these levels of granularity and given the current large static component of the current systems, when the goal is to reduce power consumption, the best strategy is to reduce time. Under this conditions malleability is key. Malleability may come from a programming model like OmpSs,  and from libraries such as the Load Balancing Library (DLB). OmpSs decouples computation (tasks) from the resources (threads). DLB is able then to dynamically change  the number of threads associated with each MPI process.

In the talk we will describe the Load Balancing Library DLB and a balancing algorithm, LeWI, that can improve the performance of hybrid MPI/OmpSs applications.

The talk will be illustrated with practical examples.

Biography : Rosa M. Badia holds a PhD on Computer Science (1994) from the Technical University of Catalonia (UPC). She is a Scientific Researcher from the Consejo Superior de Investigaciones Científicas (CSIC) and team leader of the Grid Computing and Cluster research group at the Barcelona Supercomputing Center (BSC). She was involved in teaching and research activities at the UPC from 1989 to 2008, where she was an Associated Professor since year 1997. From 1999 to 2005 she was involved in research and development activities at the European Center of Parallelism of Barcelona (CEPBA). Her current research interest are programming models for complex platforms (from multicore, GPUs to Grid/Cloud). The group lead by Dr. Badia has been developing StarSs programming model for more than 10 years, with a high success in adoption by application developers. Currently the group focuses its efforts in two instances of StarSs: OmpSs for heterogenoeus platforms and COMPSs for distributed computing (i.e. Cloud). Dr Badia has published more than 120 papers in international conferences and journals in the topics of her research. She has participated in several European projects, for example BEinGRID, Brein, CoreGRID, OGF-Europe, SIENA, TEXT and VENUS-C, and currently she is participating in the project Severo Ochoa (at Spanish level), TERAFLUX, OPTIMIS, EU-Brazil OpenBIO, ScalaLife and TransPlant and it is a member of HiPEAC2 NoE.