Introduction
Our motivation originated from a prior study illustrating the negative impact of resource-intensive requests on overall application performance. We desired to create a library that could detect performance degradation, monitor runtime resource usage, select appropriate cancellation targets, and issue timely cancellations
Responsibility
- Created Go library for detecting performance degradation in etcd, a distributed key-value store
- Established lightweight runtime profiler monitoring scheduler waiting time and mutex usage of requests
- Implemented heuristic algorithm to identify criminal requests adversely affecting system performance
- Developed robust cost-benefit analysis policies to determine the cancellation of trigger requests