Sarvesh S. Kulkarni

A dynamic load dispersion algorithm for load-balancing in a heterogeneous grid system

The ever changing demands on computational resources has information systems managers looking for solutions that are more flexible. Using a ldquobigger boxrdquo that has more and faster processors and permanent storage or more random access memory (RAM) is not a viable solution as the system usage patterns vary. In order for a system to handle the peak load adequately, it will go underutilized much of the time. A grid based distributed system can solve this problem by allowing multiple independent jobs to run over a network of heterogeneous computers. Applications can be based on several parallel jobs, several sequential jobs, or a single job. Keeping the workload represented by these jobs balanced over the network of computers requires network-aware scheduling algorithms that are dynamic, transparent, scalable, and quick. We present such an algorithm that handles load-balancing of jobs submitted at any point in the grid. Our algorithm accommodates jobs with differing CPU and I/O requirements and load-balances them over varying grid loads and varying network latencies.

Design and implementation of a framework for monitoring patients in hospitals using wireless sensors in ad hoc configuration.

Patients in hospital intensive care units (ICUs) have to be monitored constantly. Typically, unless the attending physician is at the bedside, he or she has no information of the patients progress, and must be paged manually in case of emergencies. A far better system would be one that keeps the patients information available to the doctor or nurse at all times, possibly through the use of a handheld device. The doctor would then be able to check on the patients progress in real time. Should an emergency arise, the doctor would be notified directly by the monitoring system itself, saving valuable response time. This paper aims to set up an infrastructure for monitoring patients in hospital ICUs. An architectural framework and related protocols to support communication between the various components of such a system are presented. The prototyped product was successfully demonstrated to industry representatives, in November 2005, at Villanova University, PA

Making Health Care More Accessible to Rural Communities in Waslala, Nicaragua Using Low-Cost Telecommunications

Two years ago, at the first IEEE Global Humanitarian Technology Conference, we presented a paper on the development of a tele-health project to provide a communication link to health care providers for the rural communities surrounding Waslala, a town located in the North Atlantic Autonomous region of Nicaragua. The system employed volunteer community health workers (CHWs) to transmit vital signs of their community members to a computer server using low-cost SMS telecommunications technology. At that time the project was still under development and four CHWs started to work on the project. Since then considerable progress has been made with many challenges overcome along the way. We now have fifty of the 92 communities covered by the program and expect to provide full coverage to this region by next year. We are also starting to consider other rural locations in Nicaragua to expand the project. This paper elaborates on the many challenges that have been faced and overcome on this project, the key partnerships that have been established to support the program, the impact of the project to date, and the plans for project expansion over the next year.

SMART: statistically multiplexed adaptive routing technique for ad hoc networks

This paper develops a common solution to the problems of discovery, maintenance, and use of multiple routes in ad hoc networks. The performance criterion is the average time taken by a packet to reach its destination through multiple hops. A source node considers each of its neighbors (reachable by direct wireless transmission) as a next-hop for every possible destination. The effect of delay at a next-hop and beyond, until the packet reaches its destination, is approximately modeled as an equivalent M/M/1 queuing system. Available neighbors at every node provide multiple routes. Multiple routes are statistically multiplexed to distribute the load as well as to deal with changes in data rates and network configuration. The potential of each next-hop neighbor of a node in providing a viable route is estimated on-line and the proportions of traffic routed through these multiple neighbors are also updated adaptively.We study this approach and conduct extensive experiments over a network with two extreme cases of simulated traffic patterns, the Poisson, and the self-similar types. Even when the network topology is static, our algorithm responds to bursts in the traffic pattern and reduces buffer losses through the use of alternative, less congested routes. We also present simulation experiments and results to demonstrate the effectiveness of our algorithm in the presence of mobility, using self-similar traffic. Mobility is simulated by means of the random waypoint model in which nodes move with varying speeds. Results show that our simple unified approach handles the problems of mobility as well as network congestion very well.

A Poisson Based Bursty Model of Internet Traffic

This paper presents a model for bursty data traffic on modern networks. The model is simple and amenable to analysis, thus enhancing its practical utility. We use the model to generate arrivals to a single queue and compute its performance measures. We also propose a measure of burstiness based upon these performance parameters that is easy to measure and use. Several traffic traces were downloaded from the internet and analyzed to determine their burstiness properties. We conclude that predictions from the model are consistent with the data on the internet and the degradation of all the parameters due to burstiness.

A scalable architecture for performance measurement in broadband networks

Given the competitive nature of the broadband network market, providing better service to consumers in terms of measurably superior end-user experience is crucial for Internet Service providers (ISPs). The large scale and heterogeneity of network infrastructure pose a challenge in collecting data and in assessing performance metrics in such a system of systems. This paper presents our broad architecture called the Network Measurement and Monitoring Architecture (NMMA) and some key implementation details for the measurement of network performance metrics, their collection, aggregation, analysis and presentation. The work presented here was performed for a major ISP and complements the draft standards proposed by the Internet Engineering Task Forces working group on Large-Scale Measurement of Broadband Performance (IETF-LMAP). Our NMMA architectural framework and implementation details expand on the IETF-LMAP working groups framework and thus provide a reference implementation. Although not a direct concern of the proposed LMAP draft framework, our implementation also provides guidance for a uniform visual presentation of data collected from the network, for analysis purposes.

Adaptive control of heterogeneous ad hoc networks

Heterogeneous ad hoc networks are proposed as a solution to the scalability problem in pure ad hoc networks. Some of the links in heterogeneous networks have a much longer wireless transmission range (in one or more of their channels) than the other links. We study simple approaches for adaptive operation and performance optimization of such networks. Our study includes the following: (a) Use of very simple functions of observed performance for adaptive control; (b) Accommodation of an operational constraint imposed by a higher layer in the network. In this case, traffic from a forwarding node to a neighboring node, meant for an eventual destination, can be multiplexed over either long range links or over short range links but not both; (c) Route selection using extremely simple, predetermined criteria; (d) Performance comparison of the above three simplified approaches to the original algorithm of Reference [4]. The original algorithm is computationally more accurate, but handicapped by the absence of long range links. Our approaches are illustrated with simulation experiments on heterogeneous ad hoc networks fielding synthetic self-similar (bursty) traffic. The performance comparisons of the approaches are presented for both fixed as well as dynamic network topologies. They emphasize the advantage of heterogeneous network operation over a pure ad hoc network, all nodes in the latter being functionally identical. Copyright