Designing Energy Efficient Communication Runtime Systems for Data Centric Programming Models

Abstract

The insatiable demand of high performance computing is being driven by the most computationally intensive applications such as computational chemistry, climate modeling, nuclear physics, etc. The last couple of decades have observed a tremendous rise in supercomputers with architectures ranging from traditional clusters to system-on-a-chip in order to achieve the petaflop computing barrier. However, with advent of petaflopplus computing, we have ushered in an era where power efficient system software stack is imperative for execution on exascale systems and beyond. At the same time, computationally intensive applications are exploring programming models beyond traditional message passing, as a combination of Partitioned Global Address Space (PGAS) languages and libraries, providing one-sided communication paradigm with put, get and accumulate primitives. To support the PGAS models, it is critical to design power efficient and high performance one-sided communication runtime systems.In this paper, we design and implement PASCoL, a high performance power aware one-sided communication library using Aggregate Remote Memory Copy Interface (ARMCI), the communication runtime system of Global Arrays. For various communication primitives provided by ARMCI, we study the impact of Dynamic Voltage/Frequency Scaling (DVFS) and a combination of interrupt (blocking)/polling based mechanisms provided by most modern interconnects. We implement our design and evaluate it with synthetic benchmarks using an InfiniBand cluster. Our results indicate that PASCoL can achieve significant reduction in energy consumed per byte transfer without additional penalty for various one-sided communication primitives and various message sizes and data transfer patterns.

Cuttlefish: Library for Achieving Energy Efficiency in Multicore Parallel Programs

“…The approaches that target the Message Passing Interface (MPI) applications mainly involve mitigation of workload imbalance between the process (slack) [4,16,29,41]. Other MPI-centric solutions address cases where the processor cores wait on the memory or network [5,22,26,43,48,49]. Concurrency throttling has been widely used by adapting the thread count in OpenMP programs that are memory-constrained to reduce power consumption [12,13,31,37].…”

Runtime power-aware energy-saving scheme for parallel applications

“…The more sophisticated ones scale processor frequency on different intervals of application runtime while attempting to predict accurately the performance effects from the DVFS. Such approaches may be broadly classified into two types: One that first divides the application into execution intervals of predefined duration and then uses the performance counters to determine a suitable frequency for them [7,10,11]; and the other that first determines communication intervals in parallel applications that use either explicit message passing [6,15,22,23] or global address-space primitives [24] and then scales the frequency for those intervals, usually based on the variation of the MIPS (million instructions per second) metric at different P-states. Typically these approaches first choose a (often user-defined) performance loss (PL) tolerance for the application and then try to maximize energy savings under this PL as constraint.…”

Evaluating Effects of Application Based and Automatic Energy Saving Strategies On NWChem

“…The other approaches primarily focus on scaling processor frequency during slack or communication operations during application runtime. The techniques in the past have targeted communication intervals in parallel applications that use either explicit message passing Lowenthal 2005, Lim, Freeh, andLowenthal 2006) or global address-space primitives (Vishnu, Song, Marquez, Barker, Kerbyson, Cameron, and Balaji 2010) and then scales the frequency for those intervals. Oversubscribing the processor cores (Iancu, Hofmeyr, Blagojevic, and Zheng 2010) is another technique which can be used to reduce execution time and lower power consumption of a parallel application.…”