With increasing adoption of high-performance Arm-based CPUs beyond mobile devices, it is important for developers to understand Arm’s performance characteristics for common server-side software stacks. In this article, we will use AWS’s Arm (Graviton2) and x86_64 (Intel) EC2 instances to evaluate computational performance across different software runtimes, including Docker, Node.js, and WebAssembly. Our conclusion is that Arm is more cost effective in the cloud, especially with lightweight runtimes that are close to the underlying operating system.
Background
In a recent research paper published in Science, MIT professors Leiserson and Thompson et al. discussed one of today’s most important challenges in computer engineering — the end of Moore’s Law. Computer hardware, such as the CPU and GPU, had hit the quantum limit and can no longer be made much faster or smaller. That threatens 40 years of productivity and economic growth powered technology innovation. Is the technology revolution as we know it over? Yet, the paper is optimistic about our technology future.
The authors suggest that software improvements could replace Moore’s Law and drive productivity growth in the years to come. To illustrate this point, they noted that re-writing machine learning algorithms from Python to C / native code could improve performance by 60,000 times!
However, we cannot just give up modern software runtimes and the developer productivities they bring, go back to pre-Java days of the 1990’s, and run every application in compiled native code. Today’s developers rely on high-level programming languages, tooling, and especially memory safe and portable runtimes, to deliver high-quality software products.
According to the Science paper authors, the approaches to software performance engineering are to remove software bloat and tailor software to more efficient hardware.
In this article, we will evaluate the performance gains by adopting lightweight and efficient software and hardware infrastructure in cloud computing scenarios. Specially, we run several lightweight software runtimes on both energy-efficient Arm-based CPUs (AWS Graviton2) and Intel x86 CPUs.
#performance #benchmark #arm #cloud
