On understanding the energy consumption of ARM-based multicore servers

Abstract

There is growing interest to replace traditional servers with lowpower multicore systems such as ARM Cortex-A9. However, such systems are typically provisioned for mobile applications that have lower memory and I/O requirements than server application. Thus, the impact and extent of the imbalance between application and system resources in exploiting energy efficient execution of server workloads is unclear. This paper proposes a trace-driven analytical model for understanding the energy performance of server workloads on ARM Cortex-A9 multicore systems. Key to our approach is the modeling of the degrees of CPU core, memory and I/O resource overlap, and in estimating the number of cores and clock frequency that optimizes energy performance without compromising execution time. Since energy usage is the product of utilized power and execution time, the model first estimates the execution time of a program. CPU time, which accounts for both cores and memory response time, is modeled as an M/G/1 queuing system. Workload characterization of high performance computing, web hosting and financial computing applications shows that bursty memory traffic fits a Pareto distribution, and non-bursty memory traffic is exponentially distributed. Our analysis using these server workloads reveals that not all server workloads might benefit from higher number of cores or clock frequencies. Applying our model, we predict the configurations that increase energy efficiency by 10% without turning off cores, and up to one third with shutting down unutilized cores. For memory-bounded programs, we show that the limited memory bandwidth might increase both execution time and energy usage, to the point where energy cost might be higher than on a typical x64 multicore system. Lastly, we show that increasing memory and I/O bandwidth can improve both the execution time and the energy usage of server workloads on ARM Cortex-A9 systems.

A Study on the Energetic Viability of Single Board Computers for Cloud Computing Scenarios

“…From an energy-based standpoint, and extending on the outstanding analysis of Tudor [20], we have generalized the energy studies to n-machine cluster structures. This has clearly shown the performance degradation caused by the linear scale of the underlying support hardware (ram bus speed, network, storage), which will hit on the system's performance under I/O Intensive workloads, as is the case with sequential database scans.…”

A performance study of big data on small nodes

“…More generally, the related work can be classified in two categories: energy-proportionality studies [19,17,26,10] and building blocks for energy-efficient servers [8,16,31,24,27], as we further present.…”

“…As a result of the fast-evolving landscape of mobile hardware, and in a bid to reduce the energy-related costs, many companies and research projects are increasingly looking at using non-traditional hardware as server platforms [27,31]. For example, Barcelona Supercomputing Center is looking at using ARM-based systems as the basis for their exascale platform [23].…”

“…The bulk of related work can be broadly classified in two main categories: energy-proportionality studies, which aims to improve the correlation between server power consumption and the server utilization [19,17,10,26], and building blocks for energy-efficient servers, where various software and hardware techniques are used to optimize bottleneck resources in servers [8,26,31,24,27]. As the dominating architecture in datacenters has traditionally been Intel/AMD x64, most of the studies are geared toward these types of servers.…”

“…Our analysis shows that there is no one size fits all rule regarding the efficiency of the two server systems: sometimes ARM systems are better, other times Intel systems are better. Surprisingly, deciding which workloads are more efficient on small nodes versus traditional systems is not primarily affected by the bottleneck hardware resource, as observed in high-performance computing and traditional server workloads [27]. Instead, software immaturity and the challenges imposed by limited RAM size and bandwidth are the main culprits that compromise the performance on current generation ARM devices.…”