Reducing network contention with mixed workloads on modern multicore, clusters

Abstract

Multi-core systems are now extremely common in modern clusters. In the past commodity systems may have had up to two or four CPUs per compute node. In modern clusters, these systems still have the same number of CPUs, however, these CPUs have moved from singlecore to quad-core and further advances are imminent. To obtain the best performance, compute nodes in a cluster are connected with high-performance interconnects. On nearly all clusters, the number of network interfaces is the same on current multi-core systems as in the past when there were fewer cores per node. Although these networks have increased bandwidth with the shift to multicore, there still exists severe network contention for some application patterns.In this work we propose mixed workload (non-exclusive) scheduling of jobs to increase network efficiency and reduce contention. As a case-study we use Message Passing Interface (MPI) programs on the InfiniBand interconnect. We show through detailed profiling of the network that accesses of the network and CPU of some applications are complementary to each other and lead to increased network efficiency and overall application performance improvement. We show improvements of 20% and more for some of the NAS Parallel Benchmarks on quad-socket, quad-core AMD systems.

Pattern-Based Systematic Task Mapping for Many-Core Processors

“…The task mapping problem on a multi-core cluster was considered in a previous study [13]. This mapping improves performance for allocating multiple applications on a single node and uses the locality of the network.…”

Enabling fair pricing on HPC systems with node sharing

“…Previous research shows that homogeneous MPI processes can degrade one another's performance by more than 2x [20,38]. In addition, these works show that introducing heterogeneity in workloads by co-locating multiple MPI programs on disjoint cores can drastically improve performance and energy efficiency.…”

“…While this approach prevents jobs from different users from clobbering one another, it leads to a missed performance opportunity. In fact, recent work has shown that co-location, where a set of jobs from different users run on a shared set of compute nodes, can increase mean application performance and system energy efficiency by 20% by reducing contention for shared resources in the memory subsystem and inter-node network [38,33,20]. In addition, current architectural trends and exascale computing studies suggest that the benefit of co-location is likely to increase.…”

Enabling Fair Pricing on High Performance Computer Systems with Node Sharing

“…Previous research shows that homogeneous MPI processes can degrade one another's performance by more than 2× [11,33]. In addition, these works show that introducing heterogeneity in workloads by colocating multiple MPI programs on disjoint cores can drastically improve performance and energy efficiency.…”

“…on a shared set of compute nodes, can increase mean application performance and system energy efficiency by 20% by reducing contention for shared resources in the memory subsystem and inter-node network [11,27,33]. In addition, current architectural trends and exascale computing studies suggest that the benefit of co-location is likely to increase.…”