Environmentally Assured Data Design
Energy consumption correlates directly to efficient application and data design. Get this right, and you’ll save money and protect the planet.
A customer wanted to know why consumption costs were critically high for their new database cluster. The infrastructure had a high specification, but the CPU usage hit 100% shortly after the application started. It permanently stayed at this level:
Allocating more resources dropped the usage but didn’t solve the problem. CPU usage went down, but still averages around 52% utilisation:
Excessive CPU usage? All the time? That cannot be right. It uses 1KW of energy 24 hours a day, every day. They are consuming the power to run the workload and also to cool the room.
What is going on?
Investigations revealed that application users demand rapid access to their data and would not tolerate waiting for answers. The architects include a feature which continually rebuilds those application views on the chance the user might want to open them.
Querying this methodology, it became clear that the data architecture and data model were secondary thoughts compared to the application itself. The brute-force solution delivered the required performance levels at the huge expense of exceptionally high energy consumption.
How to solve it:
Data was not a primary factor during application design. Excessive energy consumption results from inefficient design and inappropriate mitigation.
Environmentally (and financially), we must reduce that energy usage, remove the direct impact, and reduce second-hand cooling costs.
Poor query planning is often the culprit in similar engagements, requiring performance tuning, data model improvements, and access methodology updates.
We reviewed the data access patterns, in combination with the data model, proposing updates to structure and query layers. We removed the constant CPU usage problem, instead providing rapid access to data as the user requested it, satisfying on-demand requirements and eliminating a significant environmental and financial cost issue.
Who can help?
The subject domains are data architecture, data modelling, performance tuning, and application development. Find a data expert who can inform your design at the right point in your application lifecycle.
About the Author:
Jamie Steele is a Azure Data Architect here at Version 1.
