Σεμινάριο Δικτύων Επικοινωνίας - Scaling Laws for Joint Content Replication and Delivery in Wireless Networks

Τίτλος: Scaling Laws for Joint Content Replication and Delivery in Wireless Networks

Ομιλητής: Savvas Gitzenis (Center for Research and Technology (CERTH)-ITI, GR)

Ημερομηνία: 24/05/2012

Ώρα: 16:00

Αίθουσα:  Aίθουσα Συνεδριάσεων A56, Τμήματος Πληροφορικής και Τηλεπικοινωνιών

Σύνοψη:

The scalability of wireless networks has gained significant interest since the seminal work of Gupta&Kumar that derived the pessimistic result that in large wireless networks with multihop communications where each user uniformly selects another node to communicate with, per-user throughput falls asymptotically as to the inverse of the square root of N, the number of nodes in the network. The intuitive explanation behind this law is that to reach the destination, messages should be relayed over a number of hops that is on average proportional to the square root of N, i.e., the network diameter.

In this talk, we switch perspective to the novel paradigm of Content-based Networking and study the new resultant scaling laws. In particular, in Content-based Networking, requests are placed for specific content, as opposed to a specific node, and, moreover, nodes are equipped with buffers/caches where the content is replicated; thus, the requested content may exist in multiple places in the network. Intuitively then, we aspire in breaching the previous O(N^.5) network diameter hop-count, and switch to more favorable scaling laws for the user throughput.

To this end, we start by formulating the detailed joint content replication and delivery problem, which is a hard combinatorial optimization. This is reduced to a simple replication density problem, that can be solved directly using KTT conditions/Lagrange multipliers. We show that the latter problem's performance is of the same order to the original problem, which enables us studying its scaling behavior. Assuming the Zipf popularity distribution regarding the content requests, well-noted in Internet traffic, we derive the associated scaling laws. In particular, the asymptotics are considered with respect to the number of nodes N of the network, volume of the content or number of files M, and the node cache capacity K. Letting them scale to infinity in various regimes, we derive laws ranging from O(N^.5), as in Gupta&Kumar, down to O(1), i.e., the best possible result.

Parts of this work have been presented in the INFOCOM 2012 and SPASWiN/WiOpt 2012 conferences.

Σύντομο Βιογραφικό:

Savvas Gitzenis has received in MSc and PhD degrees from the department of Electrical Engineering of Stanford University and his diploma/BS degree from the National Technical University of Athens. He is currently a researcher at the Institute of Informatics and Telematics of CERTH and has been involved in several FP7 research projects. Savvas has served as an adjunct lecturer in the Department of Computer and Communication Engineering at the University of Thessaly and previously has been a post doctoral researcher at Sun Labs in Menlo Park, California. His research interests lie in the area of resource allocation and distributed control in networks and computing systems, and in particular, in wireless networks and mobile computing, investigating on autonomous and scalable policies.