You’re running a SaaS business and customers are lining up to get your product. The more workloads you run, the more money you bring in. The only problem is those workloads aren’t free and figuring out the optimal hosting and reservation strategy on AWS can be a nightmare.

Getting this problem right is a delicate dance and it takes two to tango: engineering and finance need to work hand-in-hand to ensure that the reservation is a good fit for the workload and is cost-optimized to ensure the highest possible gross margin.

In this post we go through the process of how we analyzed possible strategies, presented  options, and ultimately found the most efficient solution for a SaaS customer looking to scale up their core EC2 workload in the most cost effective manner.

CPU vs. Storage vs. Memory

The first step was figuring out which dimension the workload was bound on - CPU, storage, or memory. It’s rare for a workload to be equally bound on all dimensions. Once you know what the limiting factor is, you can use that information to start your decision making process.

Our infrastructure analysis page, breaks down your net costs along various dimensions, including utilized CPU, storage and memory per machine, which can be a big help in determining your workload cost drivers

In the case we are focusing on, the customer knew that the workload in question was storage-bound. They had been using vanilla i3 instances for this workload but were curious if they could get a better deal by switching to i3en.  It was a promising approach because i3en offers the best deal on storage of any EC2 instance, but it’s also more expensive for memory and CPUs compared to vanilla i3, so it was important to look at the exact numbers. Every penny they could save would be a huge win because they were looking to massively scale up this workload.

The relative On-Demand cost per unit in the target region.

Not only is i3en more efficient for storage, but it’s a big difference, offering savings of nearly 50% compared to i3. Of course, we needed to make sure the memory and CPU needs were covered as well. We found that if the workload requires less than two virtual CPUs and 25 GB of memory per TB of storage, it’s almost certainly worth it to switch. The workload in question fit the bill, so i3en was looking like a great option.

Reservations

Compared to On-Demand pricing, Reserved Instances (RIs) can offer savings of up to 75%, but the RI savings rate for any particular instance can vary a lot. To find the best deal, it was important to flesh out the RI savings as well.

Savings compared to On-Demand pricing in the target region

It turned out when we considered RI savings the case for i3en got even stronger. The savings for reserving i3en were about 5% higher across the board on top of the already preferable On-Demand deal.

RIs have a few options in terms of reservation length and money upfront. Given that storage was the limiting factor, it made sense to think about the available reservations in terms of the cost per TB of storage. The next chart compares i3 and i3en instances across all the reservation options.

Knowing that the savings for i3en vs. i3 were similar across the board, the customer was free to choose their RI options based on their own level of priority and certainty for this workload.

Conclusion


With the RI savings wrapped in, switching to i3en instances promised savings of more than 50%, a tremendous opportunity for the customer to scale up with optimal gross margins.

While storage was the dominant concern for this workload, Reserved.ai specializes in creating detailed reports for workloads of all types. We present the information in a clear, easy-to-understand way so your engineering and finance leaders can have a productive conversation about the right hosting & reservation strategy for you.

You could give yourself a headache and eat up loads of time trying to figure out all your workload constraints & reservation options - or you could let our data scientists & cloud billing experts do it for you while you focus on creating more great products to scale up.