Showing papers by "Nagarajan Kandasamy published in 2013"

Datacenters as Controllable Load Resources in the Electricity Market

Abstract: Data centers, being major consumers of power, can play an important role in the efficient operation of electrical grids. This paper develops an optimization framework to allow data centers to operate as controllable load resources within the demand dispatch regime, a demand response (DR) program in which incentives are designed to induce lower electricity use not just during times of high prices but also when the reliability of the local grid is jeopardized or when the electricity supply and demand are unbalanced. Assuming the availability of geographically distributed and virtualized data centers situated in multiple regional electrical markets, the basic idea is to migrate the workload in the form of virtual machines (VMs) between these centers to maximize the expected payoff. The proposed framework addresses issues specific to the demand dispatch of data centers such as timeliness of VM migrations and the impact of geographic distance on migration times. It also explicitly incorporates risks that may cause the load curtailment operation to be ultimately unsuccessful and result in monetary losses to data center operators; specifically, variability in network bandwidth that can cause uncertainty in VM migration times as well as the uncertain payoff when participating in DR markets. A set of case studies involving datacenters participating in an economic DR program is used to validate the framework.

High Performance Deformable Image Registration Algorithms for Manycore Processors

Abstract: High Performance Deformable Image Registration Algorithms for Manycore Processors develops highly data-parallel image registration algorithms suitable for use on modern multi-core architectures, including graphics processing units (GPUs). Focusing on deformable registration, we show how to develop data-parallel versions of the registration algorithm suitable for execution on the GPU. Image registration is the process of aligning two or more images into a common coordinate frame and is a fundamental step to be able to compare or fuse data obtained from different sensor measurements. Extracting useful information from 2D/3D data is essential to realizing key technologies underlying our daily lives. Examples include autonomous vehicles and humanoid robots that can recognize and manipulate objects in cluttered environments using stereo vision and laser sensing and medical imaging to localize and diagnose tumors in internal organs using data captured by CT/MRI scans. This book demonstrates: How to redesign widely used image registration algorithms so as to best expose the underlying parallelism available in these algorithms How to pose and implement the parallel versions of the algorithms within the single instruction, multiple data (SIMD) model supported by GPUs Programming "tricks" that can help readers develop other image processing algorithms, including registration algorithms for the GPU

Deformable Registration Using Optical-Flow Methods

Abstract: Optical-flow methods describe the registration problem as a set of flow equations, under the assumption that image intensities are constant between views. The most common variant is the “demons algorithm,” which combines a stabilized vector-field estimation algorithm with Gaussian regularization. The algorithm is iterative, and alternates between solving the flow equations and regularization. We describe data-parallel designs for the demons deformable registration algorithm, suitable for use on a GPU. Streaming versions of these algorithms are implemented using the CUDA programming environment.