Nagarajan Kandasamy

Power and Performance Management of Virtualized Computing Environments Via Lookahead Control

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, 26% of the power required by a system without dynamic control while still maintaining QoS goals.

GPU-based streaming architectures for fast cone-beam CT image reconstruction and demons deformable registration

This paper shows how to significantly accelerate cone-beam CT reconstruction and 3D deformable image registration using the stream-processing model. We describe data-parallel designs for the Feldkamp, Davis and Kress (FDK) reconstruction algorithm, and the demons deformable registration algorithm, suitable for use on a commodity graphics processing unit. The streaming versions of these algorithms are implemented using the Brook programming environment and executed on an NVidia 8800 GPU. Performance results using CT data of a preserved swine lung indicate that the GPU-based implementations of the FDK and demons algorithms achieve a substantial speedup--up to 80 times for FDK and 70 times for demons when compared to an optimized reference implementation on a 2.8 GHz Intel processor. In addition, the accuracy of the GPU-based implementations was found to be excellent. Compared with CPU-based implementations, the RMS differences were less than 0.1 Hounsfield unit for reconstruction and less than 0.1 mm for deformable registration.

On developing B-spline registration algorithms for multi-core processors.

Spline-based deformable registration methods are quite popular within the medical-imaging community due to their flexibility and robustness. However, they require a large amount of computing time to obtain adequate results. This paper makes two contributions towards accelerating B-spline-based registration. First, we propose a grid-alignment scheme and associated data structures that greatly reduce the complexity of the registration algorithm. Based on this grid-alignment scheme, we then develop highly data parallel designs for B-spline registration within the stream-processing model, suitable for implementation on multi-core processors such as graphics processing units (GPUs). Particular attention is focused on an optimal method for performing analytic gradient computations in a data parallel fashion. CPU and GPU versions are validated for execution time and registration quality. Performance results on large images show that our GPU algorithm achieves a speedup of 15 times over the single-threaded CPU implementation whereas our multi-core CPU algorithm achieves a speedup of 8 times over the single-threaded implementation. The CPU and GPU versions achieve near-identical registration quality in terms of RMS differences between the generated vector fields.

Risk-Aware Limited Lookahead Control for Dynamic Resource Provisioning in Enterprise Computing Systems

Utility or on-demand computing, a provisioning model where a service provider makes computing infrastructure available to customers as needed, is becoming increasingly common in enterprise computing systems. Realizing this model requires making dynamic and sometimes risky, resource provisioning and allocation decisions in an uncertain operating environment to maximize revenue while reducing operating cost. This paper develops an optimization framework wherein the resource provisioning problem is posed as one of sequential decision making under uncertainty and solved using a limited lookahead control scheme. The proposed approach accounts for the switching costs incurred during resource provisioning and explicitly encodes risk in the optimization problem. Simulations using workload traces from the Soccer World Cup 1998 web site show that a computing system managed by our controller generates up to 20% more revenue than a system without dynamic control while incurring low control overhead.

Self-optimization in computer systems via on-line control: application to power management

Computer systems hosting critical e-commerce applications must typically satisfy stringent quality-of-service (QoS) requirements under dynamic operating conditions and workloads. Also, as such systems increase in size and complexity, maintaining the desired QoS by manually tuning the numerous performance-related parameters will become very difficult. This paper addresses the design of self-optimizing computer systems using a generic online control framework in which the control actions governing the operation of the system are obtained by optimizing its behavior, as forecast by a mathematical model, over a limited time horizon. As a specific application of this control technique, we show how to minimize the power consumed by a single computer processing a time-varying workload. Assuming a processor capable of operating at multiple frequencies, we design an online controller to satisfy the QoS requirements of the workload while operating the processor at the lowest possible frequency. We describe the processor model, formulate the power management problem, and derive the online control algorithm. The performance of the controller is evaluated using representative e-commerce workloads. Finally, we discuss how the proposed technique can be applied to other resource management problems in computer systems.

Online control for self-management in computing systems

Dependable computer systems hosting critical commerce, transportation, and military applications, among others, must satisfy stringent quality-of-service (QoS) requirements. However, as these systems become increasingly complex, maintaining the desired QoS by manually tuning the numerous performance-related parameters are very difficult. This paper develops a generic online control framework to design self-managing computer systems. The proposed approach explores a limited region of the system state-space at each time step and decides the best control action accordingly. We present two case studies to demonstrate the practicality of the proposed control framework.

A control-based framework for self-managing distributed computing systems

This paper describes an online control framework to design self-managing distributed computing systems that continually optimize their performance in response to changing computing demands and environmental conditions. An online control technique is used in conjunction with predictive filters to tune the performance of individual system components based on their forecast behavior. In a distributed setting, a global controller is used to manage the interaction between components such that overall system requirements are satisfied.

Enabling Self-Managing Applications using Model-based Online Control Strategies

The increasing heterogeneity, dynamism and uncertainty of emerging DCE (Distributed Computing Environment) systems imply that an application must be able to detect and adapt to changes in its state, its requirements and the state of the system to meet its desired QoS constraints. As system and application scales increase, ad hoc heuristic-based approaches to application adaptation and self-management quickly become insufficient. This paper builds on the Accord programming system for rule-based self-management and extends it with model-based control and optimization strategies. This paper also presents the development of a self-managing data streaming service based on online control using Accord. This service is part of a Grid-based fusion simulation workflow consisting of long-running simulations, executing on remote supercomputing sites and generating several terabytes of data, which must then be streamed over a wide-area network for live analysis and visualization. The self-managing data streaming service minimize data streaming overheads on the simulations, adapt to dynamic network bandwidth and prevent data loss. An evaluation of the service demonstrating its feasibility is presented.

On the application of predictive control techniques for adaptive performance management of computing systems

This paper addresses adaptive performance management of real-time computing systems. We consider a generic model-based predictive control approach that can be applied to a variety of computing applications in which the system performance must be tuned using a finite set of control inputs. The paper focuses on several key aspects affecting the application of this control technique to practical systems. In particular, we present techniques to enhance the speed of the control algorithm for real-time systems. Next we study the feasibility of the predictive control policy for a given system model and performance specification under uncertain operating conditions. The paper then introduces several measures to characterize the performance of the controller, and presents a generic tool for system modeling and automatic control synthesis. Finally, we present a case study involving a real-time computing system to demonstrate the applicability of the predictive control framework.

Risk-aware limited lookahead control for dynamic resource provisioning in enterprise computing systems

Abstract Utility or on-demand computing, a provisioning model where a service provider makes computing infrastructure available to customers as needed, is becoming increasingly common in enterprise computing systems. Realizing this model requires making dynamic, and sometimes risky, resource provisioning and allocation decisions in an uncertain operating environment to maximize revenue while reducing operating cost. This paper develops an optimization framework wherein the resource provisioning problem is posed as one of sequential decision making under uncertainty and solved using a limited lookahead control scheme. The proposed approach accounts for the switching costs incurred during resource provisioning and explicitly encodes risk in the optimization problem. Simulations using workload traces from the Soccer World Cup 1998 web site show that a computing system managed by our controller generates up to 20% more profit than a system without dynamic control while incurring low control overhead.