Compute Capacity Management
In private cloud environments, capacity management is a necessary evil. Cloud resources are not without cost, but they can seem that way to end-users. Zero capacity means “the cloud is broken!” to some, which could decelerate cloud adoption within companies. In this talk we will provide real-world examples on how eBay Inc has significantly reduced waste within our own cloud. Mitigating or Eliminating Waste - High-level Strategies
- Smart Reclamation using Telemetry or other tools (Cloud Minion) -and/or-
- Show-back -and/or-
A deep look at what criteria and metrics can be collected and baselined against to identify underutilized compute. Tooling Reporting Framework
- General strategies around engaging with waste owners.
- Tell the users, and tell their managers also, when you detect compute that may be underutilized
- Don’t stigmatize - these aren’t abusers, they are enthusiastic users - this is a good problem to have and a fun one to solve.
- Use a “lease” concept to ask users to renew their instances when low utilization is detected
- Cycles to consider: report, halt, delete
- Delay between cycles: general thoughts and best practices
eBay Inc’s Cloud Minion tool has been used to reduce compute waste for several months with great success, reclaiming (redacted) dollars worth of otherwise wasted compute. Ask the developers and experts at eBay Inc questions on how we accomplished this, and gain a deeper understanding of the open source Cloud Minion project.