| Since the publication of Gupta and Kumar paper [1] in March 2000, it has been known in the research community that building very large ad hoc networks is a challenge; although the total throughput of a flat-routed network increases with the number of nodes, the per-connection end-to-end throughput decreases. In the limit, the per-connection end-to-end throughput approaches 0. Furthermore, the standard engineering approach of “divide and conquer,” where the network nodes are partitioned into clusters, and where the communication among clusters is conducted through a high-tier network, does not resolve the vanishing end-to-end throughput either. It seems that given some communication spectrum, and at some values of communication parameters, to achieve some end-to-end throughput, one is required to keep the number of nodes below some level. The ugly face of scalability curse shows up!
In this paper, we introduce our approach to circumventing this scalability curse. Our approach is based on a two-tiered architecture, whereas the lower tier comprises of “clouds” of standard RF-based MANETs, which are then interconnected by the Free Space Optics (FSO) technology.
The results of our work teach us how RF-based MANET could be grown to large sizes (i.e., number of nodes), while preventing the per node end-to-end throughput to vanish. To achieve this, the network area needs to grow as well with the increase in the number of nodes, and such a growth in the area needs to follow specific patterns, which we have derived in our work. Then, we use the FSO technology which, relying on spatial reuse, allows us to practically indefinitely increase the size of the network, without affecting its end-to-end performance.
The main rationale behind integration of low tier peer-to-peer RF network with the higher tier peer-to-peer FSO network stems from the fact that the two technologies are complementary. In fact, the particular requirements of the lower tier network are well satisfied by the RF technology, while the particular constraints of the higher tier network well match the FSO technology.
[1] P. Gupta and P. R. Kumar, “The capacity of wireless networks,” IEEE Trans. Inform. Theory, vol. 46, no. 2, pp. 388–404, March 2000
|
| Professor Zygmunt J. Haas (wireless and mobile networks, wireless communication systems) is the director of the Cornell's Wireless Network Laboratory (WNL). The WNL (wnl.ece.cornell.edu) has been extensively involved in research into the various aspects of the Ad Hoc Networking Technology in areas such as: routing, MAC design, security, multicast, scalability, topology control, etc. The ad hoc networking community has exhibited strong interest in the WNL publications and, in particular, in the release of the second version of WNL's scalable network simulator (JiST) that is capable of simulating hundreds of thousands of nodes, orders of magnitude larger than any other tool publicly available until now. Indeed, JiST's performance significantly advances the state-of-the-art simulation capability. In the area of stochastic routing, WNL has proposed and developed algorithms for "gossiping," a technique that allows highly efficient coverage of very large networks. These practical low-complexity algorithms are implementable with efficiency close to the theoretical bounds. The WNL's first paper on gossiping in wireless networks was the eighth most cited paper in the year 2003 according to CiteSeer, the Scientific Literature Digital Library. Prof. Haas's research group also continues to study a number of biologically-inspired networks whose operations (e.g., routing, topology control) are inspired by phenomena from the biological world. WNL researchers have shown that animal mobility could be used to implement "delay-tolerant" networks with considerable efficiency by trading storage for delay and delay for energy. |