Bikash Agarwalla

DFuse: a framework for distributed data fusion

Simple in-network data aggregation (or fusion) techniques for sensor networks have been the focus of several recent research efforts, but they are insufficient to support advanced fusion applications. We extend these techniques to future sensor networks and ask two related questions: (a) what is the appropriate set of data fusion techniques, and (b) how do we dynamically assign aggregation roles to the nodes of a sensor network. We have developed an architectural framework, DFuse, for answering these two questions. It consists of a data fusion API and a distributed algorithm for energy-aware role assignment. The fusion API enables an application to be specified as a coarse-grained dataflow graph, and eases application development and deployment. The role assignment algorithm maps the graph onto the network, and optimally adapts the mapping at run-time using role migration. Experiments on an iPAQ farm show that, the fusion API has low-overhead, and the role assignment algorithm with role migration significantly increases the network lifetime compared to any static assignment.

Dynamic data fusion for future sensor networks

DFuse is an architectural framework for dynamic application-specified data fusion in sensor networks. It bridges an important abstraction gap for developing advanced fusion applications that takes into account the dynamic nature of applications and sensor networks. Elements of the DFuse architecture include a fusion API, a distributed role assignment algorithm that dynamically adapts the placement of the application task graph on the network, and an abstraction migration facility that aids such dynamic role assignment. Experimental evaluations show that the API has low overhead, and simulation results show that the role assignment algorithm significantly increases the network lifetime over static placement.

Streamline: scheduling streaming applications in a wide area environment

Scheduling a streaming application on high-performance computing (HPC) resources has to be sensitive to the computation and communication needs of each stage of the application dataflow graph to ensure QoS criteria such as latency and throughput. Since the grid has evolved out of traditional high-performance computing, the tools available for scheduling are more appropriate for batch-oriented applications. Our scheduler, called Streamline, considers the dynamic nature of the grid and runs periodically to adapt scheduling decisions using application requirements (per-stage computation and communication needs), application constraints (such as co-location of stages), and resource availability. The performance of Streamline is compared with an Optimal placement, Simulated Annealing (SA) approximations, and E-Condor, a streaming grid scheduler built using Condor. For kernels of streaming applications, we show that Streamline performs close to the Optimal and SA algorithms, and an order of magnitude better than E-Condor under non-uniform load conditions. We also conduct scalability studies showing the advantage of Streamline over other approaches. Furthermore, we implement Streamline on Planetlab as a grid service and demonstrate that it performs close to SA algorithm under dynamic resource conditions.

Streamline: a Scheduling Heuristic for Streaming Applications on the Grid

Scheduling a streaming application on high-performance computing (HPC) resources has to be sensitive to the computation and communication needs of each stage of the application dataflow graph to ensure QoS criteria such as latency and throughput. Since the grid has evolved out of traditional high-performance computing, the tools available for scheduling are more appropriate for batch-oriented applications. Our scheduler, called Streamline, considers the dynamic nature of the grid and runs periodically to adapt scheduling decisions using application requirements (per-stage computation and communication needs), application constraints (such as co-location of stages), and resource availability. The performance of Streamline is compared with an Optimal placement, Simulated Annealing (SA) approximations, and E-Condor, a streaming grid scheduler built using Condor. For kernels of streaming applications, we show that Streamline performs close to the Optimal and SA algorithms, and an order of magnitude better than E-Condor under non-uniform load conditions. We also conduct scalability studies showing the advantage of Streamline over other approaches.

Architecture for resource allocation services supporting interactive remote desktop sessions in utility grids

Emerging large scale utility computing systems like Grids promise computing and storage to be provided to end users as a utility. System management services deployed in the middleware are a key to enabling this vision. Utility Grids provide a challenge in terms of scale, dynamism, and heterogeneity of resources and workloads. In this paper, we present a model based architecture for resource allocation services for Utility Grids. The proposed service is built in the context of interactive remote desktop session workloads and takes application performance QoS models into consideration. The key design guidelines are hierarchical request structure, application performance models, remote desktop session performance models, site admission control, multi-variable resource assignment system, and runtime session admission control. We have also built a simulation toolkit that can handle mixed batch and remote desktop session requests, and have implemented our proposed resource allocation service into the toolkit. We present some results from experiments done using the toolkit. Our proposed architecture for resource allocation services addresses the needs of emerging utility computing systems and captures the key concepts and guidelines for building such services in these environments.

A system architecture for distributed control loop applications (extended abstract)

In this position paper we motivate an important emerging class of applications that cooperate across a complex distributed computational fabric containing elements of widely varying capabilities, including physical and virtual sensors, actuators, and high-performance computational clusters and grids. We identify typical requirements of such applications and identify several novel research challenges that such applications pose. We sketch an evolving architecture developed as part of the Media Broker project at Georgia Tech that solves a subset of the problems presented.

Resource allocation for remote desktop sessions in Utility Grids

Emerging large‐scale utility computing systems such as Grids promise computing and storage to be provided to end users as a utility. System management services deployed in the middleware are a key to enabling this vision. Utility Grids provide a challenge in terms of scale, dynamism and heterogeneity of resources and workloads. In this paper, we present a model‐based architecture for resource allocation services for Utility Grids. The proposed service is built in the context of interactive remote desktop session workloads and takes application performance QoS models into consideration. The key design guidelines are hierarchical request structure, application performance models, remote desktop session performance models, site admission control, multi‐variable resource assignment system and runtime session admission control. We have also built a simulation framework that can handle mixed batch and remote desktop session requests, and have implemented our proposed resource allocation service into the framework. We present some results from experiments using the framework. Our proposed architecture for resource allocation services addresses the needs of emerging utility computing systems and captures the key concepts and guidelines for building such services in these environments. Copyright

Performance study of a cluster runtime system for dynamic interactive stream-oriented applications

Emerging application domains such as interactive vision, animation, and multimedia collaboration need specialized runtime systems that provide support mechanisms to enable plumbing, cross module data transfer, buffer management, synchronization and so on. Using Stampede, a cluster programming system that is designed to meet the requirements of such applications, we quantify the performance of such mechanisms. We have developed a timing infrastructure that helps tease out the time spent by an application in different layers of software, viz., the main algorithmic component, the support mechanisms, and the raw messaging. Several interesting insights have surfaced from this study. First, memory allocation does not take up a significant amount of the execution time despite the interactive and dynamic nature of the application domain. Second, the Stampede runtime adds a minimal overhead over raw messaging for structuring such applications. Third, the results suggest that the thread scheduler on Linux may be more responsive than the one on Solaris. Fourth, the messaging layer spends quite a bit of time in synchronization operations. Perhaps the most interesting result of this study is that general-purpose operating systems such as Linux and Solaris are quite adequate to meet the requirements of emerging dynamic interactive stream-oriented applications.