- While most servers spend the majority of their time well below peak usage, companies often pay for max usage 24/7.
- Cloud providers enable the ability to scale usage up and down but determining the right schedule is highly prone to human error.
- Machine learning models can be used to predict server usage throughout the day and scale the servers to that predicted usage.
- Depending on the number of servers, savings can be in the millions of dollars.
How big of a server do you need? Do you know? Enough to handle peak load, plus a little more headroom? How often is your server going to run at peak utilization? For 2 hours per day? 10 hours? If your server is only running at 2 hours per day at peak load, then you are paying for 22 hours of peak performance that you aren’t using. Multiply that inefficiency across many servers, and that’s a lot of money spent on compute power sitting idle.
Cloud Providers Make Scaling Up and Down Possible (with a Caveat)
If you’ve moved off-premise and are using a cloud provider such as AWS or Azure, it’s easy to reconfigure server sizes if you find that you need a bigger server or if you’re not fully utilizing the compute, as in the example above. You can also schedule these servers to resize if there are certain times where the workload is heavier. For example, scheduling a server to scale up during nightly batch processes or during the day to handle customer transactions.
The ability to schedule is powerful, but it can be difficult to manage the specific needs of each server, especially when your enterprise uses many servers for a wide variety of purposes. The demands of a server can also change, perhaps without their knowledge, requiring close monitoring of the system. Managing the schedules of servers becomes yet another task to pile on top of all of IT’s other responsibilities. If only there was a solution that could recognize the needs of a server and create dynamic schedules accordingly, and do so without any intervention from IT. This type of problem is a great example for the application of machine learning.
How Machine Learning Can Dynamically Scale Your Server Capacity (without the Guesswork)
Machine learning excels at taking data and creating rules. In this case, you could use a model to predict server utilization, and then use that information to dynamically create schedules for each database.
Server Optimization In Action
Recently, I did such an application for a client in the banking industry, leading to a 68% increase in efficiency and a cost savings of $10,000 per year for a single server. When applied to the client’s other 2,000 servers, this method could lead to savings of $20 million per year!
While the actual savings will depend on the number of servers employed and the efficiency at which they currently run, the cost benefits will be significant once the machine learning server optimization model is applied.
If you’re interested in learning more about using machine learning to save money on your server usage, click here to learn more about our risk-free server optimization whiteboard session.