Cloudability logo
Call For Papers Now Open - Propose a Talk For CloudyCon 2019
Optimization

Amazon EC2 A1: Is It Worth It?

By Gavin Cahill on February 5, 2019
The A1 family represents a whole new direction for EC2 instances, one with Arm architecture and AWS-designed silicon. But can it deliver the cost savings it promises?

Released at AWS re:Invent 2018, the A1 family is the first EC2 instance family built around the Arm architecture — and the first family to use chips designed by AWS. A1 is a general purpose family, along with the M families and T families, but A1 instances have a lower cost per CPU than other general purpose instance families. AWS promises significant cost savings for the right workloads. So what are those workloads? And more importantly, can A1 instances deliver on that promise and save you money?

EC2 A1 Specs & Price

A1 instances are powered by AWS Graviton Processors with 64-bit Arm Neoverse cores. Like other EC2 instances from the same generation, A1 instances are powered by the AWS Nitro System, a combination of hardware and hypervisor.

Size vCPU Mem (GiB) Storage (GiB) Network Performance (Gbps) On-Demand Price
a1.medium 1 2 EBS-Only Up to 10 $0.0255/hr
a1.large 2 4 EBS-Only Up to 10 $0.051/hr
a1.xlarge 4 8 EBS-Only Up to 10 $0.102/hr
a1.2xlarge 8 16 EBS-Only Up to 10 $0.204/hr
a1.4xlarge 16 32 EBS-Only Up to 10 $0.408/hr
A1 instance specs. Prices are for Linux instances in US West (Oregon).

The key to taking full advantage of A1 instances is matching up the right workload. The Arm architecture is designed to use less energy and to have less raw compute power. This makes them a good match for lower-compute applications like web servers, containerized microservices or caching fleets. Of course, if your code is already compiled for Arm architectures, then A1 instances are a pretty strong bet. For example, A1 is a great replacement for Raspberry Pi-type server farms.

That being said, A1 instances aren’t limited strictly to Arm-compiled code. Because they run with Linux, they can use a lot of open-source software made to run on multiple processors, including Perl, PHP, Ruby, Python, NodeJS and Java.

How does A1 compare to other instances?

The A1 family is general purpose, so let’s look at how it stacks up with M5, the core general purpose EC2 family. But we’re also going to look at the C5 family, since A1’s distribution of vCPUs and memory has similar ratios to C5.

At first glance, price immediately stands out as the biggest differentiator.

Size vCPU Mem (GiB) Storage (GiB) On-Demand Price
a1.2xlarge 8 16 EBS-Only $0.204/hr
m5.2xlarge 8 32 EBS-Only $0.384/hr
c5.2xlarge 8 16 EBS-Only $0.34/hr
Prices are for Linux instances in US West (Oregon).

The A1 is ~40% cheaper to run than either the M5 or the C5, and a large part of this is the lower power consumption from the Arm architecture. After all, the hardware behind A1 instances is cheaper to run, which means AWS can charge less to use them. But it’s important to remember that the vCPUs on A1 instances aren’t as powerful as the Intel-powered vCPUs on other instances. According to benchmark tests, price for performance changes from workload to workload. Sometimes the A1 rates better, while sometimes M5 rates better.

So can A1 instances save you money?

With the right workload, A1 instances can instantly give you significant savings by migrating workloads from other EC2 instance families. But it’s absolutely crucial to get the workload right. A price that’s 40% less won’t help you if the lower compute power of the A1 means a process takes twice as long as it would on an M5 or C5.

While there are generalized approaches for finding the right workloads, the best method will always be to analyze your actual cloud usage compared to the pricing and capabilities of A1 instances. Cloudability’s rightsizing feature is designed for this exact reason. Pulling your actual usage data, it uses machine learning to create optimization recommendations, then presents them based on the clipping risk with each selection.

Showing how migrating from m5 to a1 instances saves money.

Recommendation to migrate from an m5.large instance to an a1.large and save 47%.

If you look at the above screenshot, you’ll see that the CPU usage comes in spikes. At the same time, the memory requirements are steadily pretty low. That combination means that the load could be switched to A1 with very little risk of affecting performance.

All that being said, we’ve seen slow adoption of A1 instances at Cloudability over the last couple months, accounting for less than 1% of EC2 instances we’re monitoring. But we’ll definitely be watching to see where this goes in the future as companies start optimizing their workloads and creating more use cases that take full advantage of the Arm architecture.

——

Want to find out more about choosing the right EC2 instance? Download our Choosing the Right EC2 Instances to Optimize Your Cloud e-book!

Being in the know feels great