Please note: This PhD seminar will be given online.
Brad Glasbergen, PhD candidate
David R. Cheriton School of Computer Science
Supervisor: Professor Khuzaima Daudjee
Data systems are known for their complexity; they contain a vast number of features and configuration parameters to support different use cases. As no single data system can efficiently process all workload types, administrators face a daunting task:
- They must continuously monitor their systems for performance degradations and failures.
- When issues arise, they must rapidly comb through large debug log files and hundreds of exported system metrics to understand the problem's origin and fix it.
- They must deploy and monitor multiple data systems, each specialized for particular aspects of their overall workload.
In this talk, I show how to alleviate this administrator burden by bolting an intelligent monitoring and adaptivity framework onto existing data systems. This framework, Dendrite, extracts models of system behaviour with low-overhead from built-in debug logs, detects shifts in behaviour, and deploys customizable adaptations to respond to changes in client workloads. I will demonstrate Dendrite’s utility through practical case studies with popular data systems and empirical evaluation.
To join this PhD seminar on Zoom, please go to https://uwaterloo.zoom.us/j/94349250988.
200 University Avenue West
Waterloo, ON N2L 3G1