Combinatorial Optimization Done Right: How to Schedule Networking Tasks and Deploy In-Network Processing Services in an Optimal Fashion
Matthias Rost
Technische Universität Berlin
Friday, May 30
2:00 p.m., Conference Room 6A
Virtualization of node and link resources has been the key for enabling flexible resource allocations both in data centers and across wide-area networks. In this talk we will present our latest results in this field:
- We introduce a novel continuous-time model for the scheduling of networking tasks. As our model allows for arbitrarily fine-granular allocations several real-world problems can be modeled. We outline our model and present an initial computational evaluation showing that, e.g., wide-area communications between data centers can be efficiently scheduled.
- We propose a novel unifying model for deploying in-network processing services. While our model accurately models multicasting and geo-replication, it can also be applied to compute (approximate) solutions for aggregation-type tasks like in-network analytics.
We highlight the importance of using the "right" optimization approach - and show how to obtain it - and present both exact and heuristic algorithms for solving the respective problems.
Bio:
Matthias Rost joined Anja Feldmann's Group at Technische Universität Berlin as undergraduate in 2011 and has become a PhD candidate in the same group in January 2014. Matthias’ main research focus lies on solving algorithmic problems that arise with the introduction of virtualization into the networking layer. For more information see http://matthias-rost.de/