Navigating EC2 pricing - the better way
When it comes to choosing the ideal AWS EC2 instance for your workload, navigating through the myriad of options can be a daunting task. Factors like memory, CPU, instance type, and family interplay in a complex pricing 💵 dance 🕺 that isn’t always logical (at least not on the surface), often leaving you puzzled about the most cost-effective choice 😕
To simplify this process for myself I went ahead and developed an interactive plot that maps out the relationship between AWS EC2 instance types, their associated memory, CPU configurations, and pricing in the AU ap-southeast-2 region. This tool allows me to make informed decisions by visualizing the cost-effectiveness of different instance types based on their specific needs.
How often doing bioinformatics dev work do you say I just need a couple of CPUs, but a good 💪 amount of RAM, maybe 64GB, but what if I max that out 🤔 should I just go for more at the start, say 96GB? I wonder whether that will cost much extra? What even instance family do I want for this?…Sure go ahead trawl 🔍 through the endless AWS tables and lists of EC2 instance type, compare, then search for the hourly price, cross reference… or just look at a plot 👇
The x-axis represents memory allocation while the y-axis denotes the hourly price. I chose memory or RAM to be be my x-axis as I feel it is most commonly the more important determining variable in a bioinformatic dev environment. Number of vCPUs is represented by the various colours and CPU clock speed (GHz) controls the point size. Each point correspond to a specific instance type available for EC2 on-demand usage, and full instance details are displayed when hovering over the point, revealing additional information such as network and storage configurations.
Note
To keep the plot focused and useable I removed instances with more than 1000GB of RAM and/or more than 72 vCPUs. It is also unlikely you wouldn’t be after these for a dev environment - they are better suited to BATCH queues with multi job processing.
Warning
These prices were valid in October 2023, however, they are likely to change overtime.
It pays to shop around
While there is not surprisingly a correlation between memory and pricing, there are scenarios across the spectrum where you can get an instance with more RAM 🐏 at a lower price than another instance with less RAM, and the same with vCPU number, etc. It is about finding those sweet spots 🍭 that work for our needs.
An example of this: say you are after an instance with 16GB of RAM you might go for the standard compute optimised c5.2xlarge for $0.44/hr, orrr you could instead get a memory optimised z1d.xlarge with a higher clock speed and twice as much RAM (32GB) for pretty much the same price $0.452/hr 😊️ orrr the memory optimised r5a.xlarge also with 32GB RAM but at a much lower $0.272/hr 😀 Yes nearly 40% cheaper than maybe your initial default choice but with twice as much RAM 😏👍
Future Prospects:
As of now, the interactive scatter plot encompasses data solely from the AU ap-southeast-2 region. I’m open to expanding and updating this tool to include additional regions or further refine its functionalities based on the community’s interest and feedback. Some of things features I am thinking of adding are:
the ability to highlight instances with similar prices, when hovering over a point
the ability to search for a instance type/name
add jitter to prevent overlapping points - the ability to copy the instance name when hovering