Sandeep Kakumanu

Research Projects

Multichannel Wireless Networks

1. Component Based Channel Assignment in Single Radio Multi-channel Ad-Hoc Networks (March 2005 - present):

In this project, we consider the channel assignment problem in single radio multi-channel mobile ad-hoc networks. Specifically, we investigate the granularity of channel assignment decisions that gives the best trade-off in terms of performance and complexity. We present a new granularity for channel assignment that we refer to as component level channel assignment . The strategy is relatively simple, and is characterized by several impressive practical advantages including (i) no changes to the off-the-shelf radio hardware or MAC algorithms, (ii) no synchronization requirements, (iii) no channel scheduling overheads, and (iv) no switching between channels to serve data flows. We use ns-2 simultations and a testbed setup to show the advantages of the proposed startegy. We also show that the theoretical performance of the component based channel assignment strategy does not lag significantly behind the optimal possible performance, and perhaps more importantly we show that when coupled with its several practical advantages, it significantly outperforms other strategies under most network conditions.,

Publications:
[1] 'Component Based Channel Allocation in Single Radio Multi-channel Ad-Hoc Networks', Ramanuja Vedantham, Sandeep Kakumanu, Sriram Lakshmanan and Raghupathy Sivakumar, MOBICOM 2006, LA.
[2] 'Component Based Channel Assignment (CBCA): A New Channel Assignment Strategy for Multi-Channel Single-Interface Ad-hoc Networks', Sandeep Kakumanu, Ramanuja Vedantham, Sriram Lakshmanan and Raghupathy Sivakumar, Submitted to Transactions on Networking (TON).

2. Lattice Routing: A 4D Routing Scheme for Multiradio Multichannel Adhoc Networks (may 2007 - Dec 2007):

An efficient channel assignment strategy ensures capacity maximization in a multiradio, multichannel adhoc network. Existing mechanisms either use a static channel assignment or a centralized process intensive system that assigns channels to individual nodes. These are not effective in a dynamic environment with multiple flows that are active at different time instants. The protocol proposed in this work manages channels of the radios for the different nodes in the network using information about current channel conditions and adapts itself to varying traffic patterns in order to efficiently use the multiple channels. Further the protocol uses multipathing, a key mechanism that is found to alleviate bottlenecks present in single path routes in such an environment. ns-2 based simulations indicate the benefits of the proposed protocol when compared and contrasted with related work. This work was part of my summer internship at Los Alamos National Laboratory during summer of 2007.

Publications:
[1] 'Lattice Routing: A 4D Routing Scheme for Multiradio Multichannel Adhoc Networks' Sandeep Kakumanu, Stephan Eidenbenz, and Raghupathy Sivakumar, Under Submission.

3. Hydra: Enabling Multi-Radio 802.11 Communication for High Data-Rate Wireless Networks (Oct 2007 - present):

In this paper, we consider the following question: Using only off-the-shelf 802.11 radios, what is the maximum wireless communication data rate that can be achieved between two devices? Through test-bed experiments that use all 12 possible 802.11a channels (5.2Ghz band), we find that although the ideal expected application-level throughput is approximately 500Mbps, the actual performance achieved is only 65Mbps. We use a combination of wireline benchmark analysis, wireless packet trace analysis, and spectrum analysis to identify the reasons for the low performance. We then present Hydra, a solution that overcomes the aforementioned reasons and delivers close to ideal expected aggregate performance. We evaluate the performance of Hydra using a prototype implementation on a 12-radio testbed built with atheros based 802.11 cards.

Other Projects

4. VoIP Call Capacity over IEEE 802.11 Networks (June 2006 - July 2006):

The expected VoIP call capacity ia a one-hop IEEE 802.11b network with the G.711 voice codec is about 85 simultaneous calls, but the actual observed capacity is only 5 calls even at the highest rate and under zero loss conditions. In this work we analyze the reasons behind this inferior performance of VoIP traffic. We also present algorithms at the medium access layer to improve the observed call capacity. Finally, using ns-2 based simulations, we evaluate the algorithms, and show that performance improvements in the order of upto 400% can be achieved.

Publications:
[1] 'Improving call capacity of VoIP over 802.11 networks,' Yeonsik Jeong, Sandeep Kakumanu, Cheng-Lin Tsao, and Raghupathy Sivakumar, IEEE Broadnets 2007.
[2] 'Improving call capacity of VoIP over 802.11 networks,' Yeonsik Jeong, Sandeep Kakumanu, Cheng-Lin Tsao, and Raghupathy Sivakumar, Invited for publication in special issue of ACM MONET 2008.
Patent:
[1] 'Improving call capacity of VoIP over 802.11 networks,' Raghupathy Sivakumar, Yeonsik Jeong, Sandeep Kakumanu, and Cheng-Lin Tsao. Patent Filed in Nov.2006.

5. Enhancing IPTV using sensor networks (Oct 2006 - May 2007):

IPTV and Internet Video deployments using the public Internet, provisioned Internet, or satellite data networks is expected to have two key challenges: (i) congestion related quality degradation, and (ii) intolerable channel switching latency. In this work, we have developed techniques that use wireless sensor networks intelligently to address these two problems. First, using a wireless sensor network to detect the presence of a user watching the TV, appropriate de-prioritization of streams that are not being actively seen by a user is achieved, thus delivering superior performance to the other streams that actually have active watchers. Second, using a wireless sensor network to detect the picking up of a TV remote, appropriate pre-fetching of “next” channels are achieved to drastically reduce the actual channel switching latency. In addition to the above applications, future applications are envisioned including targeted advertisements based on sensor network readings (e.g. advertisements for air conditioners are enabled when the wireless sensor network reports high temperatures) and user specific content migration and delivery. Unix sockets were used to send send sensor information from one machine to another. We also modified the VLC open source video client to integrate video streaming with inputs from sendor devices. A working demonstration of the concept was shown at the semi-annual GTBI-IAB (Georgia Tech Broadband Institute – Industry advisory Board) meeting held this fall (2006). Techniques for targeted advertising are currently under development.

Publications:
[1] 'Enhancing IPTV Using Wireless Sensor Networks,' Sandeep Kakumanu, Yeonsik Jeong, Sriram Lakshmanan, and Raghupathy Sivakumar, Demonstration and presentation at ACM Mobicom 2007.
[2] 'Cue-Based Networking Using Wireless Sensor Networks: A Video-Over-IP Application,' Yeonsik Jeong, Sriram Lakshmanan, Sandeep Kakumanu, and Raghupathy Sivakumar, IEEE SECON 2008.

6. Wireless P2P syetems (Oct 2006 - March 2007):

In this project we study the effects of the wireless medium and host mobility on the performance of Peer-to-Peer networks. We identify a number of non trivial challenges in such a scenario by providing insights into the problems and propose a deployable solution that can be added in an application-transparent fashion. Briefly, some of these issues include: (i) P2P applications, unlike in typical scenarios where a mobile host functions as a client, creates a scenario requiring the mobile host to function as a server. (ii) P2P data sharing, again uniquely, typically involve simultaneous bi-directional data transfer . This consequently results in the use of bi-directional TCP, a form of TCP not studied extensively for wireless environments. (iii) P2P data networks, by virtue of being almost entirely supported by end-hosts, typically use incentives based performance delivery. In other words, a mobile host that has uploaded more data is provided with higher performance. Such a mechanism exposes issues when applied as-is to a wireless and mobile environment. (iv) While incentives encourage P2P users to stare data longer, P2P data fetching is also adapted to increase the uniquely shareable data available at a user. One such approach is performing random or rarest-first fetching.

Publications:
[1] 'On the Impact of Mobile Hosts in Peer-to-Peer Data Networks,' Zhenyun Zhuang, Sandeep Kakumanu, Yeonsik Jeong, and Raghupathy Sivakumar, Submitted to IEEE ICDCS 2008.

6. Ultra-fast Multipath transport (Feb 2007 - May 2007):

This project was a part of a course (Advance Internetwork Architectures). In this project we implemented a new transport mechanism using Linear codes (network coding) over UDP. We used multiple paths over the internet for a given source destination pair. An overlay network was created using nodes from planetlab on the internet. This overlay was used for providing multipath UDP connections. The linear codes were implemented using C++. We were able to observe throughputs of up to 100Mbps over the internet from a source in Georgia Tech to a destination in University of Washington.

7. Design of a network monitoring tool on the intel IXP processor (Feb 2006 - May 2006):

This project was a part of a course (High Performance Parallel Computing). A module was developed for the intel IXP 2400 processor to monitor network traffic to a server running several virtual machines on Xen. Assembly level coding was used for the module.

8. Convergence analysis of the BGP routing protocol using large scale distributed simulations on cluster computers (Sep 2005 - Nov 2005):

This project was part of a course (Parallel and Distributed Simulations). The project involved parallelizing a serial code to simulate a number of BGP routers and study the convergence of the protocol when some routers fail. We modified the open source GTNets simulator to implement the paralle BGP code. The parallel code was tested and the convergence properties of BGP were analyzed on a cluster machine with 60 processors.

© Sandeep Kakumanu
This layout was inspired from anirudh's webpage who plagiarized it from here.