Online control for self-management in computing systems

Abdelwahed, Sherif; Kandasamy, Nagarajan; Neema, Sandeep

doi:10.1109/rttas.2004.1317283

Cited by 53 publications

(68 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We solve the above problem using limited lookahead control (LLC), a predictive control approach previously introduced in [23]. This method is quite useful when control actions have dead times, such as switching on a server and waiting for the bootstrap routine, and for control actions that must be chosen from a discrete set, such as the number of hosts and VMs to switch on.…”

Section: Problem Formulationmentioning

confidence: 99%

Power and Performance Management of Virtualized Computing Environments Via Lookahead Control

Kusic

Kephart

Hanson³

et al. 2008

2008 International Conference on Autonomic Computing

228

View full text Add to dashboard Cite

There is growing incentive to reduce the power consumed by large-scale data centers that host online services such as banking, retail commerce, and gaming. Virtualization is a promising approach to consolidating multiple online services onto a smaller number of computing resources. A virtualized server environment allows computing resources to be shared among multiple performance-isolated platforms called virtual machines. By dynamically provisioning virtual machines, consolidating the workload, and turning servers on and off as needed, data center operators can maintain the desired quality-of-service (QoS) while achieving higher server utilization and energy efficiency. We implement and validate a dynamic resource provisioning framework for virtualized server environments wherein the provisioning problem is posed as one of sequential optimization under uncertainty and solved using a lookahead control scheme. The proposed approach accounts for the switching costs incurred while provisioning virtual machines and explicitly encodes the corresponding risk in the optimization problem. Experiments using the Trade6 enterprise application show that a server cluster managed by the controller conserves, on average, 22% of the power required by a system without dynamic control while still maintaining QoS goals. Finally, we use trace-based simulations to analyze controller performance on server clusters larger than our testbed, and show how concepts from approximation theory can be used to further reduce the computational burden of controlling large systems.

show abstract

Section: Problem Formulationmentioning

confidence: 99%

Power and Performance Management of Virtualized Computing Environments Via Lookahead Control

Kusic

Kephart

Hanson³

et al. 2008

2008 International Conference on Autonomic Computing

228

View full text Add to dashboard Cite

show abstract

“…Based on our earlier work [3], the notions of uncertainty and risk will be incorporated within these frameworks to cope with unexpected changes to performance goals, dynamic workload, and hardware and software failures. Online parameter tuning and model learning techniques will need to be integrated within the control framework to both improve the quality of partially specified system models and adapt to changes in the system model itself over time (e.g., addition or removal of system components, replacement of components).…”

Section: Design and Architecturementioning

confidence: 99%

High confidence software for cyber-physical systems

Abdelwahed

Kandasamy

Gokhale

2007

Proceedings of the 2007 Workshop on Automating Service Quality: Held at the International Conference on Automated Software Engi

View full text Add to dashboard Cite

Many new and planned cyber-physical systems (CPSs) are realized as distributed real-time and embedded (DRE) systems. Examples of DRE CPSs we are interested in include data computing centers and automated warehouse management systems. Categories and Subject Descriptors General TermsAlgorithms, Management, Design, Reliability, Verification. KeywordsCyber-physical systems. Emergent Traits of DRE CPSsNext generation DRE CPSs illustrate the following emergent characteristics:• Complexity and scale. As an increasing number of services are moved online, the required computing infrastructure keeps growing in complexity and scale. Effectively managing the performance of these large-scale computing systems subsequently becomes more complex, requiring increasing numbers of skilled personnel to configure, operate, and optimize them.• Dynamic and uncertain operating environment. These systems typically execute in a dynamic and uncertain operating environment caused by multiple factors such as time-varying user workload, hardware and software resource failures, incomplete knowledge of the system operating state, and other vulnerabilities, such as security violations or denial of service attacks.• Use of on-demand computing models. This is an emerging resource provisioning model to efficiently host applications, where computing resources e.g., servers, memory, and storage for data centers or warehouse resources e.g., forklifts and belts for warehouses, are dynamically made available to these applications as needed, and not statically allocated based simply on peak demand• High-confidence is an essential quality of service (QoS) property of these emergent DRE CPSs that must be considered over the entire lifecycle of these systems, including design, development, deployment, and operation. Meeting these demands using today's stovepiped technologies is tedious, error-prone, and costly to develop, optimize, validate, deploy, and maintain. R&D Needs for DRE CPSsRealizing high-confidence software for DRE CPSs requires meeting the following criteria.• Trustworthiness -DRE CPSs must be trustworthy, which includes the system's ability to meet performance objectives, and be resilient to failures and security attacks.• Autonomicity -DRE CPSs must be autonomous i.e., self healing, self configuring and self optimizing while maintaining good resource utilization.• Analyzability -The algorithms and technologies used to develop DRE CPSs must be amenable to analyses and verification for different properties, such as timeliness guarantees, fault tolerance, stability and correctness. Design and ArchitectureThe solution needs of high confidence DRE CPSs can be met by developing novel techniques described below and synergistically integrating them as shown in Figure 1.Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first p...

show abstract

“…The OLC [11] and control theory, talked about in the previous section, is used in modeling the controllers. The model captures the system modes, the monitored and controlled parameters and the relationship between these parameters across the modes.…”

Section: Dqme Overviewmentioning

confidence: 99%

“…A model-predictive online control (OLC) approach [11] has been developed, in which control actions are derived to optimize system behavior for pre-specified QoS criteria over a limited look-ahead prediction horizon. This type of modelpredictive control method is more powerful and more widely applicable to computational systems, which may be highly non-linear, than simple feedback control.…”

Section: Introductionmentioning

confidence: 99%