Recent Research

Content cache management for delay improvement

Coding for Distributed Storage

We consider a variation of the coupon-collector problem, where users interested in a set of coupons arrive randomly in the system and leave only when they have the complete set. Coupons are stored at a set of servers, each containing only one coupon. We show that the mean number of users waiting in the system can be reduced by storing ‘‘coded’’ coupons at the servers. This study has applications for distributed storage of contents over data centers.

  • P. Parag and J.-F. Chamberland. Waiting on Distributed Content. IEEE Information Theory and Applications (ITA), San Diego, CA, February 10-15, 2013.

Distributed Content Caching

In a traditional content-distribution network, user-requests need to be directed to the caches that have the content of interest. In turn, these caches may need to periodically update their content from a central repository. There is a trade-off between size of the cache and the frequency of updates. Further, since the network has communication capacity constraints, traffic management must be done in parallel with caching. We design throughput-optimal joint caching and load-balancing policies that also yield good delay performance.

Fundamental limits on delay-sensitive communications

Coding versus Queueing, over Correlated Erasure Channels

We study the relationship between code rate selection and queueing performance in digital communication systems. For achieving capacity in wireless channels, one often needs to employ very long codes, resulting in large decoding delays. For real-time applications with finite delay constraints, one needs shorter codes for timely decoding. We study the trade-off between code rate and service guarantees, for finite length codes operating over correlated erasure channels. A rigorous framework that links code rate to overall system performance for random codes is presented. Guidelines for code rate selection in delay-constrained system are identified.

Network coding to improve delay-limited rate region

Next, a simple QoS-constrained butterfly network was considered. We identified the potential benefits of network coding in the context of delay-sensitive applications. It was found that given a structure suitable for network coding, algebraic combining of packets always outperforms classical routing. However, in case of wireless network where one needs to perform resource allocation to set-up such a structure; it may not always be advisable to do so, depending on the specific topology.

  • P. Parag and J.-F. Chamberland. Queueing analysis of a butterfly network for comparing network coding to classical routing. IEEE Transactions on Information Theory, 56(4):1890–1908, April 2010. ieeexlpore. paper.

  • P. Parag and J.-F. Chamberland. Queueing analysis of a butterfly network. IEEE International Symposium on Information Theory, 672–676, Toronto, Canada, July 6-11, 2008. ieeexlpore. paper. slides.

User cooperation to improve delay-limited rate region

A simple two-user service-constrained multiple access wireless network was considered. %The achievable peak-rate region was identified for non-cooperative users with stochastic arrivals and service guarantees. It was shown that user-cooperation improves the achievable peak-rate region even when both users have stochastic arrivals and service guarantees, provided the quality of inter-user link is better than the individual connections from the users to the destination.

  • L. Liu, P. Parag, and J.-F. Chamberland. Quality of service analysis for user cooperation in wireless communication systems using fluid models. IEEE Transactions on Information Theory, 53(10):3833–3842, October 2007. ieeexlpore. paper.

Fundamental limits on delay-limited point-to-point communication

A point-to-point communication over a varying wireless channel was considered, with stochastic arrivals and statistical service guarantees. The joint impact of spectral bandwidth, transmit power, and code rate was considered. Analytical expressions for the probability of buffer overflow and the effective capacity were obtained. Fundamental performance limits for Markov wireless channel models were identified. %The impact of channel correlation on supportable peak-rates was investigated. It was found that, even with an unlimited power and spectral bandwidth budget, only a finite arrival rate can be supported for the QoS constraint.

  • L. Liu, P. Parag, J. Tang, W.-Y. Chen, and J.-F. Chamberland. Resource allocation and quality of service evaluation for wireless communication systems using fluid models. 44th Allerton Conference on Communication, Control, and Computing, 44:1187–1193, Monticello, IL, September 27-29, 2006.

Delay improvement using network protocols

Delay-aware distributed rate adaptation protocol

For real-time applications, users may not have high throughput requirements but stringent service constraints. We develop a distributed resource allocation algorithm for arbitrary network topologies that takes into account the end-to-end dissatisfaction of a user (e.g. end-to-end delay) due to quality degradation seen by it. The algorithm maximizes total network utility, subject to the end-to-end dissatisfaction constraints. The proposed algorithm is also shown to be stable over time.

Content exchange protocol for delay reduction

We extend this framework to include peer-to-peer assisted content distribution. A peer swarm interested in a desired content, can choose between ‘‘costly and quick’’ direct download from a server and ‘‘free and delayed’’ peer-to-peer communication. We study this tradeoff for a single ISP swarm and find the optimal routing-in-time policy. We also study multiple selfish peer-to-peer swarms competing for the single server, and the inefficiency of selfishness as compared to the social optimal.

  • P. Parag, S. Shakkottai, and I.  Menache. Service Routing in Multi-ISP P2P Content Distribution: Local or Remote? 2nd International ICST Conference on Game Theory for Networks, Shanghai, China, April 16-18, 2011. springer. paper. slides.

Other Research

Quantization for Positive Random Variables with Infinite Support

Efficient quantization schemes exist for random variables with bounded support. We explore quantization at a more fundamental level, focusing on positive random variables of unbounded support. We exploit established properties of Banach spaces, together with the boundedness of probability measures, to provide bounds on the number of bits necessary to achieve a desired level of performance.

  • P. Parag and J.-F. Chamberland. Exploiting an Interplay between Norms to Analyze Scalar Quantization Schemes. IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP), Prague, Czech Republic, May 22-25, 2011. ieeexlpore. paper. slides.

Subcarrier Allocation for OFDMA

In multi-user OFDMA systems, users see diverse realizations of the multiple subcarriers. For elastic traffic, a new subcarrier allocation algorithm was proposed, that exploits this diversity gain when buffer and channel state information are available at the transmitter. The proposed algorithm was shown to improve throughput performance when compared to existing algorithms.

  • P.  Parag, S. Bhashyam, and R. Aravind. A subcarrier allocation algorithm for OFDMA using buffer and channel state information. IEEE 62nd Vehicular Technology Conference, 62 (1):622–625, September 2005. ieeexlpore. paper. slides.