Just-in-Time Capacity Planning using Machine Learning

Increasingly Cloud Service Providers and Enterprises are using Just-in-Time (JIT) Capacity Planning to operate public and private data centers. JIT capacity planning has great potential for improving margins, allowing Capex reallocation and Opex reduction, provided future demand can be forecasted accurately. Accurate future demand forecasting is a complex undertaking. It requires analyzing massive amount of historical data and human judgment. High complexity and large-scale deployment of Web applications that often lead to workload spikes makes future demand forecasting even more problematic.

 Machine Learning offers an improvement over current forecasting method. Statistical Machine Learning (SML) techniques including Regression, Anomaly Detection, and Feature Selection can be used to build accurate workload models based on the data generated by applications and underlying hardware infrastructure (compute, networking, and storage). These models can accurately predict the capacity needs for any future workload spikes. These models also empower Cloud Service Providers and Enterprises to procure capacity in smaller increments and run data centers at higher utilization - close to the actual demand. 

 Majority of software applications today are hosted in virtualized environments that generate large and rich data sets. These data sets lend very well for applying Machine Learning techniques. Techniques such as Artificial Neural Network and Support Vector Machine are perfect for modeling the performance of virtualized application. Cloud Service Providers and Enterprises can leverage these models to maximize the utilization of their data centers by precisely estimating capacity of Virtual Machines (VMs) and efficiently allocating resources to VMs. Cloud Service Providers have an opportunity to create new pricing model based on VMs performance instead of configuration. Take a Machine Learning primer course.