[Paper Review] ARM Wrestling with Big Data: A Study of ARM64 and x64 Servers for Data Intensive Workloads.

Jayanth Kalyanasundaram, Yogesh Simmhan|arXiv (Cornell University)|Jan 21, 2017

Cloud Computing and Resource Management8 references3 citations

TL;DR

This paper evaluates an ARM64 server (AMD A1100-series) against an x64 server (AMD Opteron 3300-series) for Big Data workloads using Apache Hadoop v2.7. The ARM64 system achieves comparable performance to x64 for integer-intensive tasks like Sort and Hive queries, while consuming 1/3 the energy and achieving a 50–71% lower Energy Delay Product (EDP), demonstrating strong potential for energy-efficient data centers.

ABSTRACT

ARM processors have dominated the mobile device market in the last decade due to their favorable computing to energy ratio. In this age of Cloud data centers and Big Data analytics, the focus is increasingly on power efficient processing, rather than just high throughput computing. ARM's first commodity server-grade processor is the recent AMD A1100-series processor, based on a 64-bit ARM Cortex A57 architecture. In this paper, we study the performance and energy efficiency of a server based on this ARM64 CPU, relative to a comparable server running an AMD Opteron 3300-series x64 CPU, for Big Data workloads. Specifically, we study these for Intel's HiBench suite of web, query and machine learning benchmarks on Apache Hadoop v2.7 in a pseudo-distributed setup, for data sizes up to $20GB$ files, $5M$ web pages and $500M$ tuples. Our results show that the ARM64 server's runtime performance is comparable to the x64 server for integer-based workloads like Sort and Hive queries, and only lags behind for floating-point intensive benchmarks like PageRank, when they do not exploit data parallelism adequately. We also see that the ARM64 server takes 1/3rd the energy, and has an Energy Delay Product (EDP) that is $50-71\%$ lower than the x64 server. These results hold significant promise for data centers hosting ARM64 servers to reduce their operational costs, while offering a competitive performance for Big Data workloads.

Motivation & Objective

• To evaluate the performance and energy efficiency of ARM64-based servers for data-intensive workloads in cloud environments.
• To assess whether ARM64 processors can deliver competitive performance relative to x64 processors in Big Data analytics workloads.
• To measure the energy consumption and Energy Delay Product (EDP) of ARM64 versus x64 servers under real-world Big Data benchmarks.
• To determine the suitability of ARM64 servers for deployment in data centers aiming to reduce operational costs through energy savings.

Proposed method

• Benchmarking was conducted using Intel's HiBench suite, covering web, query, and machine learning workloads.
• Experiments were run in a pseudo-distributed Hadoop v2.7 environment to simulate real data center conditions.
• Workloads included data sizes up to 20GB files, 5 million web pages, and 500 million tuples to assess scalability.
• Performance was measured in terms of runtime, energy consumption, and Energy Delay Product (EDP), calculated as energy × delay.
• The comparison focused on integer-intensive tasks (e.g., Sort, Hive queries) and floating-point-intensive tasks (e.g., PageRank).
• The ARM64 server used the AMD A1100-series based on 64-bit ARM Cortex A57 architecture, while the x64 server used the AMD Opteron 3300-series.

Experimental results

Research questions

• RQ1 How does the runtime performance of an ARM64 server compare to an x64 server for integer-based Big Data workloads such as Sort and Hive queries?
• RQ2 What is the performance gap between ARM64 and x64 servers for floating-point intensive benchmarks like PageRank when data parallelism is limited?
• RQ3 To what extent does the ARM64 server reduce energy consumption compared to the x64 server across diverse Big Data workloads?
• RQ4 How does the Energy Delay Product (EDP) of the ARM64 server compare to that of the x64 server, and what does this imply for operational efficiency?

Key findings

• The ARM64 server achieved runtime performance comparable to the x64 server for integer-based workloads such as Sort and Hive queries.
• The ARM64 server showed a performance lag for floating-point intensive benchmarks like PageRank when data parallelism was not adequately exploited.
• The ARM64 server consumed only one-third of the energy required by the x64 server across the evaluated workloads.
• The Energy Delay Product (EDP) of the ARM64 server was 50–71% lower than that of the x64 server, indicating superior energy efficiency.
• The performance and energy efficiency advantages of the ARM64 server were consistent across data sizes up to 20GB, 5 million web pages, and 500 million tuples.
• These results suggest that ARM64 servers can significantly reduce operational costs in data centers while maintaining competitive performance for most Big Data workloads.