Implications of Overcommitment on Cloud Computing Workloads
Open AccessIn an overcommitted datacenter, the sum of resource requests for all tasks running in the datacenter is generally greater than the available capacity of that datacenter. Pushing overcommitment too far can cause increased task failures – so a balance is needed between the overcommitment rate and the task failure (task eviction) rate. The conditions under which overcommitment can be pushed higher to improve utilization without worsening the task failure rate can be understood better. This Praxis creates explainable predictive models to effectively understand the impact of job arrivals, evictions and overcommitment rates on resource utilization using Generative Discreate Event Simulations. It further explores the implications of machine heterogeneity on resource (CPU/Memory) utilization in an overcommitted environment. The effect of interactions between job arrival rates and overcommitment rates in predicting eviction rates are also studied. Results indicate that an overcommitted heterogenous cluster improves Memory and CPU utilization compared to a similar homogenous cluster when job arrival rates and eviction rates are held constant. We learn that CPU and Memory utilization can be predicted using job arrival rates, overcommitment rates and eviction rates. We also find that, at constant scheduling rates, eviction rate prediction can be improved by incorporating an interaction between job arrival rates and overcommitment rates. Finally, this research shows that a generative discrete event simulation approach can be used to model an overcommitted datacenter workload.
- All rights reserved
Notice to Authors
If you are the author of this work and you have any questions about the information on this page, please use the Contact form to get in touch with us.
| Thumbnail | Title | Date Uploaded | Visibility | Actions |
|---|---|---|---|---|
|
|
Shanbhag_gwu_0075A_15677.pdf | 2022-03-06 | Open Access |
|