Cost efficiency is an important aspect for companies of all sizes and levels of maturity, where it can become a competitive advantage. Especially on large organizations composed of hundreds of teams, it is hard to enforce initiatives that require buy-in and significant implementation effort from the various teams that often have other priorities.
In this session you will see how you can implement a cost savings initiative that provides multiple easy to use options for running at the lowest possible cost, that can be leveraged at scale with the least amount of buy-in and configuration effort from the teams. The session will explore topics such as resource tagging, centralized reserved instance purchases, automated instance terminations outside office hours and automated use of the spot market with minimal configuration changes to existing environments.
In summary, we are HERE and we are making sense of the world through the lens of location
We help people achieve better outcomes – whether it’s:
A driver moving to their destination safely
A municipality managing its infrastructure smartly
A business optimizing its fleet of assets
We are deeply driven by how we can improve the future. And that future has many opportunities and challenges:
By 2050 we will have close to 10 billion people on this planet
At least 50 megacities (each with more than 10 million, some up to 40 million, inhabitants)
”Virtual roads” are needed so that drones can glide through the system
Questions on how best sustain our resources
“Aging population”. Machines will have to teach machines. Robotics will rise
What technology to make sense of the enormous amounts of data that we collect every day?
One thing is certain: Location is very important for tackling these problems
Over the next few slides I’ll talk about how we are using AWS at HERE Technologies
We have a significant footprint on AWS, maintained by our hundreds of product teams
Each product team maintains its own AWS accounts (usually one for dev and another for prod.)
We have a few teams that ensure compliance throughout the fleet (security, availability, cost control etc.)
But ultimately each team is responsible for its own infrastructure, budget, priorities and can choose the AWS services it needs
At our scale we spend a lot of money on AWS
I can’t give exact numbers but the figure is in the millions monthly
Even small percentages of cost reductions here and there can make a significant difference
We often see that a few hours of optimization work could cover someone’s salary
We implemented a number of measures over the years, I will go over them over the next slides
We prefer to implement them centrally and make them available to the teams with minimal effort, sometimes the teams can just take advantage of them, but for some they need to take some action and configure them on their infrastructure.
Basically our sourcing team contacted AWS and agreed on some discount if we keep paying as much as we do over the next few years
Requires no work from the product teams
We are continuously analyzing our usage and trying to cover as much as possible of the stable capacity
We use AWS Organizations and we can purchase RIs in one account of the smallest instance types . They can apply for any product team that and are automatically merged to cover larger instance types
Usually allocated to Production environments where the baseline capacity tends to be relatively steady
In the diagram you can see how much coverage we got to one of our production environments over a period of two weeks
That’s how a typical developer calendar looks like, not much happens before 9AM and after 6 PM
Development environments see no traffic at night or over the week-end
We can save some 70% of the total cost if we hibernate or shut them down when idle
For AutoScaling you can also set time-based policies
For EC2 we use a tool called CloudCustodian
We use an open source project which I started in my spare time a couple of years ago, called AutoSpotting
We have it installed in each of our accounts since June, the product teams just need to use it by tagging their ASGs
That’s how it looked after enabling it in one of our Production environments, we also benefit from the RI coverage
We noticed the adoption rate of the instance scheduling and AutSpotting is very low, even though it just takes minutes to set them up.
So we will need to find ways to make them ‘hands-free’ or run them by default.
AutoSpotting supports running in opt-out mode, in which it can perform the replacement action on all the groups from the AWS account, unless blacklisted by tagging or protected from termination/scale-in.