Improving the uptime of critical applications using ML-based forecasting algorithms

Introduction

The success of many organizations today relies heavily on the uptime of their critical applications. However, maintaining optimal uptime can be challenging, especially for organizations with large, complex IT environments. This case study outlines how a big data approach leveraging machine learning algorithms was used to improve the uptime of critical applications and generate significant cost savings for a customer.

Client Requirement

The client had a need for an advanced analytics solution that could proactively indicate corrective actions to improve critical applications’ uptime. The solution was expected to have alerts and key performance indicators (KPIs) updated every 30 minutes for the next 24 hours.

Solution

To meet the client’s requirements, the solution involved identifying patterns and conditions that generate high volumes of transactions and CPU usage. Machine Learning algorithms were then brought into play to predict and forecast peak load times for the various applications over the next 24-hour period.

The forecasts were scored per application (Tier 0) and hour of the day, with root cause explanations. High scores were automatically flagged, and notifications were sent out to take corrective action. The end application was a mashup of Qlik Sense, which provided the customer with real-time insights into their critical applications.

Results

The solution had a significant impact on the customer’s business. By leveraging machine learning algorithms, the customer was able to proactively identify potential issues that could affect the uptime of their critical applications. This resulted in a reduction in downtime and an increase in overall application uptime, leading to improved productivity and increased customer satisfaction.

Additionally, the solution generated cost savings for the customer, with overall savings of up to $100K per day on server allocation costs. By accurately forecasting transactions and CPU usage, the customer was able to optimize server utilization, reducing the need for additional servers and resulting in significant cost savings.