Improving Scalability of Capacity and Capability Clusters with Intelligent Fabrics
SESSION: Cluster Interconnects
EVENT TYPE: Exhibitor Forum
TIME: 4:30PM - 5:00PM
ABSTRACT: Capacity and capability computing on Linux clusters face significant, but different, scalability challenges.
Capability clusters, which enable single applications to scale to their maximum size, are constrained by communication patterns, such as MPI collective operations. Existing solutions to this challenge have had very limited success. A truly scalable solution must include intelligence and compute power across the fabric, not just on compute nodes.
Capacity clusters, which serve many workloads simultaneously, must maximize total bandwidth while minimizing the impact on runtime of any single workload. Typical solutions involving local, dynamic routing at the switch level might create larger problems in other parts of the fabric. A more effective solution takes into account end-to-end job traffic patterns to truly maximize utilization with no hidden costs.
This session will review how Voltaire’s intelligent fabric solutions address these challenges and can help you improve the scalability of your clusters.