Research :: Resource Provisioning in Virtualized Datacenters
Principal investigator: Evangelia Kalyvianaki
Co-investigator: Themistoklis Charalambous
Modern omnipresent server applications are complex programs that provide diverse services to thousands of users, e.g., Amazon and Google services. Their demanding operations require a powerful deployment base delivered from contemporary data centers equipped with hundreds or thousands of commodity machine units. In this environment, machine resource provisioning is central for each application to comply with its service level agreements and to efficiently administer shared data center machines hosting multiple applications.
Figure: Modern Enterprise Web Server. Server applications employ a multi-tier architecture to cope with the complexity of their operations. Server components span across multiple tiers and components within each tier could also be replicated to face demanding workloads. Server-side tiers are hosted within a data centre.
Commonly, disjoints sets of machines are dedicated to different applications and to ensure performance guarantees, a common practice is to over-provision applications with machines to cope with their most demanding workloads, however rare they may be. Nevertheless, over-provisioning causes machine under-utilization in modern data centers as shown by several reports, servers typically run at 15-20 % of their capacity. Furthermore, maintaining under-used machines at scales of contemporary data centers, has severe financial consequences in capital expenses and additional power costs.
To alleviate these issues, current data centers employ server consolidation - multiple applications running on the same machines -to run fewer but more utilized machines. In fact, this approach has recently gained significant attention due to the virtualization of commodity hardware that provides the basic mechanisms to ensure performance isolation and guarantees. When virtualized, a physical server is transformed into one or more virtual machines (VMs) and applications can run within these isolated environments. Each VM is subject to management operations such as creation, deletion, and migration between physical machines, as well as run-time resource allocation. These features enable resource sharing between hosted applications and physical servers and provide the mechanisms for efficient server consolidation.
To fully capitalize on these mechanisms, methods for adaptive resource allocation subject to applications' resource demands are required. Indeed, if each application is properly provisioned, unused resources can be allocated otherwise, e.g., to run additional applications. However, the application demands are difficult to estimate in advance since their characteristics typically change over time.
Figure: Virtualized prototype and control system overview. Solid lines between the controller modules and the Rubis Server System depict the MIMO controller system.
Journals:
| [1] | Evangelia Kalyvianaki, Themistoklis Charalambous and Steven Hand : "Resource Provisioning for Multi-Tier Virtualized Server Applications", In the Computer Measurement Group Journal, Spring 2010. Abstract | Full text: PDF |
Conference papers:
| [4] | Themistoklis Charalambous and Evangelia Kalyvianaki : "A Min-Max Framework for CPU Resource Provisioning in Virtualized Servers using H-infinity Filters", In The 49th IEEE Conference on Decision and Control (CDC), December, 2010. Abstract | Full text: To Be Uploaded | Poster: PDF | Presentation: PDF |
| [3] | Evangelia Kalyvianaki, Themistoklis Charalambous and Steven Hand : "Self-Adaptive and Self-Configured CPU Resource Provisioning for Virtualized Servers Using Kalman Filters", In The 6th IEEE/ACM International Conference on Autonomic Computing and Communications (ICAC 2009) , June, 2009 (Acceptance rate: 15.6 %). Abstract | Full text: PDF |
| [2] | Evangelia Kalyvianaki, Themistoklis Charalambous and Steven Hand : "Applying Kalman Filters to Dynamic Resource Provisioning of Virtualized Server Applications", In The Third International Workshop on Feedback Control Implementation and Design in Computing Systems and Networks (FeBid), June, 2008. Abstract | Full text: PDF |
| [1] | Evangelia Kalyvianaki and Themistoklis Charalambous: "On Dynamic Resource Provisioning for Consolidated Servers in Virtual Data Centers", In The Eighth International Workshop on Performability Modeling of Computer and Communication Systems (PMCCS-8), September, 2007. Abstract | Full text: PDF |
