[talks] J Jiang preFPO
Melissa M. Lawson
mml at CS.Princeton.EDU
Tue Aug 2 10:19:35 EDT 2011
Joe Wenjie Jiang will present his preFPO on Friday
August 5 at 10:30AM in Room 318B (networking lab).
Note location. His committee is: Jennifer Rexford,
advisor; Andrea LaPaugh and Augustin Chaintreau (Columbia),
readers; Mike Freedman and Mung Chiang (ELE), nonreaders.
Everyone is invited to attend his talk. His abstract
Title: Wide-Area Traffic Management for Cloud Services
Cloud services need an effective way to distribute contents across wide area networks. Providing large-scale, geographically-replicated online services presents new opportunities for coordination between server selection (to match subscribers with servers), traffic engineering (to select efficient paths for the traffic) and adaptive content caching (to place contents sources close to users). Traditional designs isolate these problems, which degrades performance, scalability, reliability and responsiveness. We leverage the theory of distributed optimization, collaborative game theory and randomized algorithms to provide solutions that jointly optimizes over different degrees of design freedom.
This dissertation proposes a sets of solutions to wide-area and data center traffic management, which consists of the following three thrusts:
(i) Cooperative content distribution and traffic engineering in an ISP network. We develop cooperation models with an increasing amount of information exchange between ISPs and content providers. We show that the lack of communication leads to sub-optimal performance, and even when extra information is used inappropriately. We propose a joint design solution to reduce such efficiency loss.
(ii) Traffic management in Nano data centers, which represents the trend of decentralizing content delivery by placing contents close to the last mile of end user connectivity and utilizing their upload capability. We solve a joint content caching and request routing problem for minimizing cross-traffic and propose a distributed implementation.
(iii) Cost-efficient data center traffic engineering with joint server placement and route selection, by exploiting multipath capability and virtual machine migrations.
Our systems can be easily implemented with incremental modifications to today's architecture such as content-selection mechanisms, and infrastructure such as switches and end hosts. The effectiveness of our schemes is demonstrated by our evaluation with real-life traffic traces.
More information about the talks