next up previous
Next: About this document ...

Manpreet Singh
4130 Upson Hall, Department of Computer Science, Cornell University, Ithaca, NY 14853.
Phone: (607) 262-0550; Email: manpreet@cs.cornell.edu
http://www.cs.cornell.edu/~manpreet

Objective
Seeking full-time position where I can enhance my systems, analytical and leadership skills while adding value to the company. I am interested in building challenging systems that can make an impact on the real world.

Summary of Qualifications

Education

Professional Experience$^*$

Computer Skills

Patents

Publications and Presentations
1
``AirGuard: A framework for troubleshooting enterprise wireless networks using desktop infrastructure'', Victor Bahl, Jitendra Padhye, Lenin Ravindranath, Manpreet Singh, Alec Wolman and Brian Zill, accepted for publication at Workshop on Hot Topics in Networking (Hotnets 2005), Nov 14-15, 2005, Maryland.

2.
``MPAT: Aggregate TCP Congestion Management as a Building Block for Internet QoS'', Manpreet Singh (speaker), Prashant Pradhan and Paul Francis, in Proceedings of IEEE International Conference on Network Protocols (ICNP 2004), Oct 5-8, 2004, Berlin, Germany.

3.
``A Generalized Framework for Network Performance Management Using End-to-end Techniques", Prashant Pradhan, Manpreet Singh, Debanjan Saha and Sambit Sahu, in the Fourth New York Metropolitan Area Networking Workshop (NYMAN 2004), Sept 10, 2004, New York.

4.
``Utilizing spare network bandwidth to improve TCP performance'', Manpreet Singh (speaker), Saikat Guha and Paul Francis, ACM Sigcomm 2005 (Work-in-Progress), Aug 22-26, 2005, Philadelphia.

5.
``Netmapper : A graph annotation service for network applications", Manpreet Singh (speaker), Prashant Pradhan, Sambit Sahu and Paul Francis, ACM Sigmetrics 2004 (Work-in-Progress), June 12-16, 2004, New York.

6.
``Heterogeneity-Aware Peer-to-Peer Multicast'', Robbert van Renesse, Kenneth P. Birman, Adrian Bozdog, Dan Dumitriu, Manpreet Singh and Werner Vogels, Proceedings of the 17th International Symposium on Distributed Computing (DISC 2003), October 2003, Sorrento, Italy.

7.
``A SLA Framework for QoS Provisioning and Dynamic Capacity Allocation'', Rahul Garg, Ramandeep Singh Randhawa, Huzur Saran and Manpreet Singh (speaker), Page(s): 129-137, Proceedings of 2002 Tenth IEEE International Workshop on Quality of Service, IWQoS 2002, May 15-17, 2002, Miami Beach, Florida.

8.
``Performance Modeling and Analysis of a Handover Channel Exchange Scheme in Mobile Cellular Communication Systems'', Manpreet Singh Dang, Amol Prakash, Rajeev Shorey, Natwar Modani and Dinesh K. Anvekar, Page(s): 665-669 vol. 2, Proceedings of 2000 IEEE International Conference on Communications, ICC 2000, June 18-22, 2000, New Orleans.

9.
``Fuzzy Logic based Handoff in Wireless Networks'', Manpreet Singh Dang, Amol Prakash, Dinesh K. Anvekar, Manika Kapoor and Rajeev Shorey, Page(s): 2375-2379 vol. 3, Proceedings of 2000 IEEE 51st Vehicular Technology Conference, VTC2000-Spring, May 15-18, 2000, Tokyo, Japan.

Professional Activities

Merit Scholarships and Awards

References

Selected Research Projects



(1) MPAT: Aggregate TCP Congestion Control as a Building Block for Internet QoS
IBM T.J. Watson Research, May 2003 - Jan 2004

Today the use of Internet QoS (i.e. diffserv or packet shaping) is generally limited to cases where the bottleneck link is under control of the organization running the application -- for instance an enterprise access link in the case of VoIP. But what if the bottleneck link is outside an organization's network? Is there any way to provide differential QoS in this case? This is a question that IBM asked in the context of their B2B Web Services systems, and which motivated my work in this area. The question has also been asked by researchers working on QoS-capable overlay networks.

In this work, we present MPAT, the first truly scalable algorithm for fairly providing differential services to TCP flows (improve performance for some flows at the expense of other flows) that share a bottleneck link. Unlike known schemes, our approach preserves the cumulative fair share of the aggregated flows even where the number of flows in the aggregate is large. Specifically we demonstrate, primarily through experiments on the real Internet, that congestion state can be shared across more than 100 TCP flows with throughput differentials of 95:1. This is up to five times better than differentials achievable by known techniques. Indeed, MPAT scalability is limited only by the delay-bandwidth product of the aggregated flows. With this tool, it is now possible to seriously explore the viability of network QoS through overlay network services. Another killer application of this work for the enterprise network is in workload management for e-business, where our system exposes the network as a managed resource to workload manager.



(2) AirGuard: Troubleshooting Enterprise Wireless Networks using Desktop Infrastructure
Microsoft Research, Mar 2005 - Aug 2005

AirGuard is a framework for troubleshooting enterprise wireless networks using desktop infrastructure. Prior proposals in this area either rely solely upon access points (APs) and mobile clients to monitor the wireless network, or augment them with dedicated sensor nodes. We believe that these approaches can be improved upon. Note that one cannot cover the entire spectrum in a comprehensive manner using only APs and mobile clients. An AP can not monitor different channels for a significant amount of time, as that would adversely impact the associated clients. Similarly, due to power constraints, mobile devices can not continuously monitor the wireless network while their users are idle. At the same time, deploying dedicated sensor nodes is an expensive proposition.

Our solution is based on two simple observations. First, in most enterprise environments, one finds plenty of desktop machines with good wired connectivity, and spare CPU and disk resources. Second, cheap USB-based wireless adapters are commonly available. By attaching USB-based wireless adapters to desktop machines, and dedicating the adapters to the task of monitoring the wireless network, we create a simple and low-cost management infrastructure. In this work, we show how the AirGuard architecture is well-suited to solving many wireless management problems including detecting unauthorized access points, handling malfunctioning APs, and performance monitoring.

We have started implementing the system, and have a small-scale deployment inside a building. We modified a wireless card driver to suit our needs and applications. We have implemented most of the tests required for detecting rogue access points, as well as some support for network performance monitoring. Our initial results from this small deployment are quite encouraging.



(3) Netmapper: A Graph Annotation Service for Networked Applications
IBM T.J. Watson Research, May 2004 - Sept 2004

Netmapper is a network mapping and annotation service for distributed applications sensitive to bandwidth availability, latency or loss. Applications (or resource managers working on their behalf) are expected to use this service to plan their network resource usage and fault response. We expose internal network state (e.g. bandwidth available on the edges) using only end-to-end experiments that are TCP-friendly. Knowledge of this internal state allows trend analysis, network performance debugging, network planning and exploitation of alternate routes, making such a service extremely useful for network providers, and also their ISP and enterprise customers.

The system takes as input the set of end-points of a distributed application, maps out the network connectivity between them, and annotates each edge (to the extent possible) with available bandwidth. Netmapper deploys end-to-end TCP probes between the end-points of an application to estimate the available bandwidth on various network paths. It then identifies the bottleneck links on these paths using a novel TCP-friendly bottleneck identification technique. The end-to-end probes are intelligently planned so that the annotation requires minimum number of probes. We have built Netmapper, and tested it out on the real Internet. We show using experiments on the real Internet that our algorithm consumes very small amount of bandwidth, and converges fast even with dynamically varying network state.



(4) Filmore: Filling Pipes with P2P Traffic, Cornell University, Oct 2004 - Mar 2005

In order to provide low-latency service, ISP backbones typically operate at only a fraction of their capacity. Not all applications, however, require low latency. In particular, most P2P applications like file-sharing, which now dominate network traffic, operate in the background. As a result, it should in theory be possible to run P2P traffic at a strictly lower priority, with far higher link utilization and latencies. This would serve to better utilize ISP infrastructure, by as much as an order of magnitude, without noticeably degrading P2P applications. The main difficulty with ``over-subscribing'' a network like this is how to keep loss rate low. In this project, we demonstrate the problem, and discuss a number of potential solutions as well as other factors like the economics of link utilization.


next up previous
Next: About this document ...
Manpreet Singh 2005-12-01