Portfolio Scheduling for Risk Management in Business-Critical Workloads Hosted by Cloud Datacenters

Large enterprises and governments shift their long-running and critical operations to run in cloud environments. The on-demand business-critical workloads raise important challenges in datacenter operation, requiring efficient online scheduling of workloads with unprecendented dynamic behavior, yet under strict service level agreements (SLAs). The topic of this work is on the resource management in clouds and especially on VM placement of business-critical workloads using portfolio scheduling, an online scheduling technique that dynamically selects a scheduling policy from a set (portfolio) by optimizing given utility functions. We focus more on utility functions which address risks in clouds of applications not meeting specific SLAs. By using risk-aware portfolio scheduling, we achieve to mitigate the severity of resource contention experienced by workloads (Operational Risk) and secure that, even if a whole datacenter fails, workloads can be absorbed by the remaining datacenters in the system (Disaster Recoverability Risk). We investigate portfolio scheduling by experimenting with real production and synthetic workloads derived from the production environment of Solvinity, a hosting provider company in the Netherlands.

Subject

portfolio scheduling
datacenter resource management
risk management
risk tolerance
operational risk
disaster recoverability risk

To reference this document use:

http://resolver.tudelft.nl/uuid:d8188239-e20e-40b1-bc0c-760aa85edb99

Part of collection

Student theses

Document type

master thesis

Rights

Files

PDF

Thesis_Report_SN_4330544.pdf

18.12 MB

Close viewer