Update in selecting the most efficient AWS EC2 instance types


We are passionate about applying math and data science to make our customers successful. We use that same intensity and focus to improve our core technology. To that end, back in May of this year, I posted an efficient frontier analysis of all Amazon Web Services (AWS) Elastic Compute Cloud (EC2) instance types to show how I assess the best technical design options for dataxu. These options are always changing as technology is always advancing. For example, earlier this week, on Nov. 7, AWS announced the general availability of the new C5 instance family.

The C5 family, which ranges from c5.large to c5.18xlarge is a compute-optimized instance family based on 3.0 GHz Intel Xeon Platinum 8000-series processors. In addition to network bandwidth and latency boosts, they claim:

“The new instances offer a 25% price/performance improvement over the C4 instances, with over 50% for some workloads. They also have additional memory per vCPU, and (for code that can make use of the new AVX-512 instructions), twice the performance for vector and floating point workloads.”

At dataxu, we’re running thousands of instances simultaneously. I’m always looking for a way to increase our system efficiency, so I decided to plug the raw numbers into my efficient frontier analysis from May 2017 and see if I agree with their claims. The updated chart is below. See the original article for how to interpret this chart.

Update Nov 2017: selecting the most efficient AWS EC2 instance types using Pareto front analysis

For key takeaways on how the C5 instance family stacks up, check out the full post on my medium page.

P3 Introduction

In addition to the introduction of the C5 family, AWS also announced the GPU optimized P3 instance family in October 2017. The P3 family shows up in the chart this month in the bottom left corner. Since my chart only credits traditional CPU and Memory the P3 instance family presents poorly, which is not an accurate representation of this machine. I don’t currently have a method to compare GPUs efficiency, if you have some thoughts on how do this properly, please reach out or leave a comment.

Interested in learning more? Read the rest on Bill’s Medium page and continue the conversation.