Abstract:
The next generation of supercomputing clusters consists of very large number of systems of multicore and multithread architecture. It is important to identify scalability constraints in existing job scheduling software as they are applied to next generation parallel systems. However, it is expensive or impossible to deploy large scale compute clusters on physical hardware and dedicate those resources to such studies. To enable such a study for the IBM LoadLeveler product, we implement a lightweight virtualization of resource agents. This allows deployment of variable sized compute clusters on a minimal number of physical machines. In this talk, we will cover the challenges of job scheduling on supercomputing and our new approaches to address some of the challenges.

Biography:
Liana L. Fong is a Senior Technical Staff Member and the manager of the Distributed System and Resource Management group at T. J. Watson Research Center. Her research activities have been in the areas of operating system structures, resource management and system performance evaluation for commercial and scientific workloads. She led numerous projects in the infrastructure and service management areas related to on demand in utility computing and grid environments. In addition, she has participated in various governmental funded supercomputing system projects such as the teraflop systems for Accelerated Strategic Computing Initiative (ASCI) and the upcoming multi-petaflop system designed under High Productivity Computing Systems (HPCS) program. Her technologies impacted multiple IBM products and resulted in many patents and publications. Working in partnership with FIU, she has been the IBM Research technical coordinator for LA Grid Initiative since its inception.