PROJECTS

MTech Thesis

"PIP-A connection-oriented multichannel TDMA-based MAC "

• Provides high throughput solution to reliable bulk data transfer in Wireless Sensor Networks.
• Uses hop-by-hop positive ACKs and end-to-end selective NACKs to conceive reliability.
• Strives to achieve high throughput by the means of driver level pipelining, multichannel usage, simplistic scheduling along end-to-end path.
• Is a first robust implementation and an extensive evaluation of a multi-hop TDMA-based wireless MAC protocol.

MTech Course Projects

" Summary Cache based Co-operative Proxies"

• Course:Computer Network
• Co-authored with Vijay Gabale
• Implementation of distributed summary cache protocol as proposed "Summary Cache: A Scalable Wide-Area Web Cache Sharing Protocol"
• Abstract: Summary Cache based proxies cooperate behind a bottleneck link & serve each others requests. This reduces overhead on the bottleneck link to the servers & alleviates user latency. Earlier approaches like ICP caused considerable inter proxy messages & used computing, memory, network resources extensively. We evaluate the performance of 'Summary based cooperating caches' by calculating number of parameters like hit ratio, false positives, false negative, network overhead & compare these with non-cooperating version. We simulate proxies on different machines & use trace file to generate user requests. Our experiments & evaluations show results in favor of 'Summary based cooperating caches'
• Report:here.

"Blind Digital Signatures, Group Digital Signatures and Revocable Anonymity"

• Course:Network Security
• Co-authored with Vijay Gabale, Ashutosh Dhekne, and Nishant Burte
• Abstract: Digital Signatures have found wide acceptance and usage due to digitization of documents and communications. Many flavors of the classical digital signature scheme are available and we present a brief study of Blind Digital Signatures and Group Digital Signatures. We also present a novel application of Group digital Signatures in the corporate world. We have implemented this application using Java Cryptographic Architecture. Our experiments show that Group Digital Signatures can be employed with relatively little overhead for large confidential documents. This can also alleviate burden off PKI by reducing number of public keys required for each individual.
• Report:here
• Presentation: here.
  

"Analysis of Routing Metrics for Wireless Mesh Networks"

• Course:Advanced Wireless Networks
• Co-authored with Vijay Gabale
• Abstract: With respect to current trend of using off-the-shelf 802.11 devices to form wireless mesh network, there are several metrics proposed for routing in mult-hop networks. In this course project, we analyse the performance of two such metrics: ETX (Expected Transmissions) and ETT (Expected Transmission Time).
  Our work is two fold. First, we characterize a link in multihop network by analysing the effects of packet size and transmisson rate over a link. We further observe how packet error rate varies with received signal strength and whether there is any correlation between the two. Second, we establish a 2-hop 4 node mesh network and quantify the behaviour of ETX and ETT in terms of delay and throughput.
  Through this study, we find that there exist a considerable correlation between packet error rate and received signal strength. We attribute the high error rate to external interference. Also, since ETT takes into account bandwidth in its value, our experiment shows that, ETT performs better in terms of throughput than ETX. We also document our experiences in using Soekris board, installing operating system and madwifi drivers, turning a node into a router and the ‘things’ to consider while performing a link level wireless experiment.
• Report:here
• Presentation: here.

"Emulation of Differentiated Services using VNUML"

• Course:Quality of Service in Computer Networks
• Co-authored with Vijay Gabale
• Abstract: Differentiated Services (DiffServ) has been proposed by IETF as a scalable QoS solution for the next generation Internet as an alternative to Integrated Services (IntServ). It has been developed for relatively simple, coarse methods of providing different levels of service for Internet traffic. The main idea is to divide traffic into a small number of classes and allocate resources on a per class basis. The Core of a diffserv network then distinguishes between small number of forwarding classes rather than individual flows as in Integrated Services. Through this project, we have implemented various components of DiffServ like classification, marking, dropping and scheduling. We have further analysed how a DiffServ enabled network behaves under varying load conditions and in terms of real time flows. For the emulation and analysis purpose, we have used VNUML (Virtual Network User Mode Linux) tool. For classification purpose, we use multi-trie approach to match IP address according to SLAs (Service Level Agreements) between service provider and customers. To do marking, we use two rate three colour marker token bucke approach and for scheduling, we employ priority scheduling with flows constrained by token buckets. Through the experiments performed, we come up with following key conclusions : (a)(b)(c)(d)
• Report:here
• Presentation: here.

"Performance Evaluation of a decentralized task scheduling algorithm"

• Course:Performance Evaluation of computer System & networks
• Co-authored with Pulkit Goyal
  
• Abstract:

  We consider the problem of mapping tasks to processor nodes at run-time (dynamic approach) in multicomputer systems. This is against static allocation where work load allocation decisions are taken before run time and tasks are assigned to individual processors of the system. In a decentralized approach, the scheduling control is distributed amongst all the nodes in the system. The scheduling strategies in such a system should incur less overhead and identify suitable remote sites for migrating extra work.
  The decentralized algorithm proposed in the paper [1], is a combination of neighborhood averaging and bidding approach and it is intended for multicomputers connected by a store and forward communication network.
  The factors that characterize the performance of the system are:
  • Topology used
  • Stability (oscillation free or not )
  • Task Trashing Level (task thrashing occurs when node spends more time in task migration than task execution)
  We intend to study the performance of the system under different topologies like mesh, hypercube and Chordal Ring, and also comment on the thrashing and stability of the system.
  For performance evaluation, the cost of migration and scheduling overhead is also taken into account. For comparison, we select random scheduling strategy which also makes use of nearest neighbor load balancing.
  

• Proposal :here
• Presentation: here

Data Mining

• Implemented following classifiers in weka to take into account attribute costs. (Co-authored with Vijay Gabale, Ashutosh Dhekne)

1. Decision Tree classifier
2. Nearest neighor classifier