In the world of modern applications, success is often defined by growth. However, as applications become more popular and user bases expand, the demands they place on databases become increasingly volatile. We’ve all faced the nightly analytics job that outgrows server capacity, or the sudden marketing push that sends traffic soaring.
For years, MongoDB Atlas has offered customers robust auto-scaling capabilities. This feature enables dedicated clusters to respond to real-time metrics and to automatically adjust resources, which is essential for maintaining performance without constant intervention. However, purely reactive systems have an inherent limitation: they only act after a threshold is crossed. In critical moments—such as during a sudden, massive spike in demand—that brief delay while resources spin up can cause user latency.
Today, MongoDB is excited to announce a fundamental shift in how MongoDB Atlas handles capacity management: Predictive Auto-Scaling for MongoDB Atlas.
Predictive Auto-Scaling anticipates scaling needs before they occur. By exploiting advanced machine learning, MongoDB Atlas now empowers both reactive scaling and proactive capacity planning—all without lifting a finger.
Moving from reaction to prediction
Contemporary auto-scaling is a vital safety net. It monitors current usage—like CPU or memory utilization—and triggers a scaling event when specific thresholds are met. While this concept works well, it relies on a signal that stress is already present in the system.
Predictive Auto-Scaling takes a more proactive approach. It utilizes sophisticated internal machine learning models to analyze a cluster’s historical operational metrics and to determine cyclical patterns and trends specific to a workload.
For example, these could include the Monday morning login rush and other predictable daily peaks. Instead of waiting for the load to hit specific triggers, MongoDB Atlas uses these insights to schedule scaling actions in advance. The system scales up the cluster in advance to ensure the necessary computational power and storage are available precisely when the application requires them.
The benefits of being prepared
Predictive Auto-Scaling is designed for applications with variable or cyclical workloads running on dedicated MongoDB Atlas clusters.
By anticipating demand, it delivers three key advantages:
Eliminate "warm-up" latency: Because resources are provisioned before traffic spikes, your users enjoy consistent, low-latency performance even during sudden surges in demand.
Optimize costs without compromise: Historically, the only way to guarantee performance during spikes was to permanently over-provision a cluster for peak loads—paying for idle resources most of the time. Predictive scaling enables organizations to provision resources in response to demand. This is particularly efficient during quiet periods, while MongoDB Atlas scales up just in time for the peak.
Set-and-forget operations: This feature integrates seamlessly with existing MongoDB Atlas auto-scaling configurations. It requires no manual scheduling or complex configuration. Boundaries are defined, and the machine learning models handle the timing.
Intelligent scaling for modern workloads
Predictive Auto-Scaling works hand in hand with regular auto-scaling mechanisms. If an unprecedented ‘black swan’ event occurs, regular auto-scaling responds immediately to protect performance. Predictive scaling ensures that for previously observed patterns that make up the bulk of an application's life, the database is always one step ahead.
This new capability is available now for MongoDB Atlas dedicated clusters. It is time to stop reacting to traffic spikes and start anticipating them.