[NRG] PhD Prospectus Defense by Ray Sweha (Tue Dec 20 @ 12:30pm, MCS-148)

Bestavros, Azer best at bu.edu
Mon Dec 19 17:50:06 EST 2011

Boston University
Computer Science Department

PhD Prospectus Defense

Tuesday December 20, 2011 / 12:30pm - 2:00pm
Location: MCS-148


Optimizing On-Demand Resource Deployment for Peer-Assisted Content Delivery

Ray Sweha
Abstract: Increasingly, content delivery solutions leverage client resources in exchange for service in a peer-to-peer (P2P) fashion. Such peer-assisted service paradigms promise significant cost reduction to providers, but suffer from the unpredictability associated with client resources, which is often exhibited as an imbalance between the contribution and consumption of resources by clients. This imbalance deprives the providers from being able to guarantee a minimum fidelity of their services. In this thesis, we propose an architectural service model that enables providers to offer higher fidelity services by leasing additional resources available through special nodes (angels) that join the service on-demand, and only if needed. Angels can be seen as nodes that produce more of the scarce resource than they consume, hence complementing resources available through clients.

Although this service model can be deployed in many settings, this thesis focuses on peer-assisted content delivery applications, in which the scarce resource is typically the uplink capacity of clients. We target three applications that require the delivery of fresh as opposed to stale content, rendering traditional caching solutions of little use. The first application is bulk-synchronous transfer, in which the goal of the system is to minimize the maximum distribution time -- the time it takes to deliver the content to all clients in a group. The second application is live streaming, in which the goal of the system is to maintain a minimal streaming quality. The third application is Tor, the anonymous onion routing network, in which the goal of the system is to boost performance (increase bandwidth and reduce latency) throughout the network while ensuring that anonymity does not fall below some minimal level.

For each of the above applications, we develop mathematical models to determine lower bounds on the angel capacity needed to meet the desirable system goals. We provide optimal constructions that meet these lower bounds by coordinating/choreographing the interactions among clients and angels. Our analytical models and optimal constructions depend on some simplifying, yet impractical,  assumptions. Thus, inspired by our models and constructions, we develop practical techniques that we incorporate into prototypical peer-assisted angel-enabled cloud services.

In the talk, I will focus on two of the three applications considered in this thesis: live streaming and anonymous routing. In the context of live streaming, I will present the design and performance evaluation of AngelCast -- a peer-assisted live streaming service -- along with plans to deploy it as a cloud service. In the context of anonymous routing, I will present the underpinnings of TorAngels -- a performance-cognizant extension of the Tor network -- along with plans to experimentally evaluate its efficacy.

Examination Committee:
Azer Bestavros (Major Advisor), John Byers, Mark Crovella, Ibrahim Matta, and Assaf Kfoury (Chair)


-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://cs-mailman.bu.edu/pipermail/nrg-l/attachments/20111219/6db78e34/attachment.html 

More information about the NRG-L mailing list