Abstract
Systems and methods are described for mitigating micro-temporal resource congestion in multi-tenant computing environments, where conventional schedulers may not effectively resolve sub-second interference. A decentralized control plane can operate at the node level, using local probabilistic profiling to generate compressed summaries of system behavior. A causal analysis engine can then calculate a correlation coefficient to attribute performance degradation in a affected workload to the resource consumption of a potential source workload. The system can handle analysis edge cases, such as resource saturation, and can employ a return-on-investment framework to select a remediation action, for example, throttling the source, boosting the affected, or relocating a workload. This approach can provide a mechanism for fine-grained performance isolation by identifying and attributing the causal sources of sub-second interference events, which may be referred to as harmonic congestion.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Singh, Ajit, "Edge-Based Causal Analysis for Remediating Micro-Temporal Resource Congestion", Technical Disclosure Commons, ()
https://www.tdcommons.org/dpubs_series/11519