|M.Sc Student||Kogan Hadas|
|Subject||Optimal Optical Router|
|Department||Department of Electrical Engineering||Supervisor||Professor Isaac Keslassy|
|Full Thesis text|
It is often claimed that future routers will necessarily be all-optical, because electronic routers are not fast enough to keep up with the increase in fiber capacity. In addition, the power consumption of electronic routers has grown exponentially and already exceeds standards.
Therefore, it seems that another solution is mandatory, and many have suggested all-optical routers as an alternative. However, two objections are commonly raised: first, basic optical components are typically very expensive; and second, optical switch reconfigurations are relatively slow. In this work, our primary goal is to find the optimal way to build an optical router using optical switches and fiber delay lines, such that as few optical components as possible are used.
We begin by establishing general results that demonstrate the cost limits of a general optical construction. We first define two fundamental costs: the first is related to the number of switch reconfigurations, and the second is related to the number of 2x2 switches in the construction. We continue by fundamentally bounding these costs, therefore stating "how good we can do". The bounds on the first cost are achieved by developing equivalence between coding theory and optical system design through introducing the concept of super switches.
We show that the minimal expected number of switch reconfigurations is almost equal to the state space entropy of the optical system. The bound on the number of fundamental optical components (2x2 switches) needed to construct a system is related back to Shannon. We point out the trade-off between the two types of costs.
After establishing general results that apply to any optical construction, we focus on the special case of an all-optical router. We construct an optical router with as few basic components as possible. First, we find a lower bound on the minimum number of basic components required to construct a router: an NxN router with a buffer size of B per port needs at least O(N log(NB)) components. Then, we present a construction that achieves this lower bound. Finally, we generalize this result also to a PIFO shared memory with buffer size NB and an OQ-FIFO switch.