Published on Nov 23, 2015
The need to transmit information in large volumes and in more compact forms is felt these days more than ever before. To provide the bandwidth necessary to fulfill the ever-increasing demand, the copper networks have been upgraded and nowadays to a great extend replaced with optical fiber networks.
Though initially these were deployed as point-to-point interconnections, real optical networking using optical switches is possible today. Since the advent of optical amplifiers allowed the deployment of dense wavelength division multiplexing (DWDM), the bandwidth available on a single fiber has grown significantly.
Optical communication can take place in one of the two ways - either circuit switching or else packet switching. In circuit switching, the route and bandwidth allocated to the stream remain constant over the lifetime of the stream. The capacity of each channel is divided into a number of fixed-rate logical channels, called circuits. Optical cross connects (OXCs) switch wavelengths from their input ports to their output ports.
To the client layer of the optical network, the connections realized by the network of OXCs are seen as a virtual topology, possibly different from physical topology (containing WDM links). To set up the connections, as in the old telephony world, a so called control plane is necessary to allow for signaling. Enabling automatic setup of connections through such a control plane is the focus of the work in the automatically switched optical network (ASON) framework.
Since the light paths that have to be set up in such an ASON will have a relatively long lifetime (typically in the range of hours to days), the switching time requirements on OXCs are not very demanding.
It is clear that the main disadvantage of such circuit switched networks is that they are not able to adequately cope with highly variable traffic. Since the capacity offered by a single wavelength ranges up to a few tens of gigabits per second, poor utilization of the available bandwidth is likely. A packet switched concept, where bandwidth is effectively consumed when data is being sent, clearly allows more efficient handling of traffic that greatly varies in both volume and communication endpoints, such as in currently dominant internet traffic.
In packet switching, the data stream originating at the source is divided into packets of fixed or variable size. In this method the bandwidth is effectively consumed when data is being sent and so allows a more efficient handling of traffic that greatly varies in both volume and communication endpoints.
In the last decade, various research groups have focused on optical packet switching (OPS), aimed at more efficiently using the huge bandwidths offered by such networks. The idea is to use optical fiber to transport optical packets, rather than continuous streams of light. Optical packets consist of a header and a payload. In an OPS node, the transported data is kept in the optical domain, but the header information is extracted and processed using mature control electronics, as optical processing is still in its infancy. To limit the amount of header processing, client layer traffic (e.g., IP traffic) will be aggregated into fairly large packets.
o unlock the possibilities of OPS, several issues arise and are being solved today. To be competitive with the other solutions, the OPS cost node needs to be limited, and the architectures should be future proof (i.e., scalable). In this context, the work of Clos on multistage architectures has been inspiring.
In the era in which Clos produced his work, the dominant telecommunications technology was analog telephony using copper wire as the transport medium. Clos’s paper helped the development of large switches: the multistage architectures proved to need far fewer cross points. This reduction in cross points was a very important issue, since it greatly determined the cost. Indeed, since Shockley, Bardeen and Brattain had only developed the transistor only a few years earlier, large scale integration as we know it today, was still a quite ambitious research topic. The main achievement of Clos was to circumvent technological boundaries, while also reducing the cost. Though communications technology has advanced much since the days of Clos, the concepts he developed are still in use.
In the context of OPS, the driving factors that lead to the adoption of multistage architectures were again reducing switch complexity (thus cost) and circumventing technological constraints. These issues are being discussed here from an architectural design point, rather than elaborating on, say, packet scheduling and routing problems in multistage switches.
One of the best known, or at least quite impressive, optical switching technologies is MEMS using tiny mirrors to deflect light from a particular input to a particular output port. Both 2D variants (where mirrors are either tilted up or lie down and let light pass) and 3D variants (with mirrors tilting along two axes) have been demonstrated. While the characteristics in terms of optical signal quality distortion are quite good, this approach is not feasible in an OPS concept where very fast switching times (range of nanoseconds) is mandatory. Two widespread approaches are one based on arrayed wave-guide grating (AWG) with tunable wavelength converters (TWCs), and another based on a broadcast-and-select (B & S) concept using, for example, SOA technology.
In both the B & S and AWG approaches, scalability issues will arise. A solution is to employ multistage architectures similar to the one proposed by Clos. The terminology being used in relation to blocking and switching can be defined as follows:
We call a switching architecture strictly nonblocking when it is always possible to connect any idle input port to any idle output port, irrespective of other connections that are already present.
A switch is rearrangable nonblocking if it is possible to connect any idle input port to any idle output port, bout some of the existing connections have to be reconfigured in order to do so. After the reconfiguration all connections are functional again.
When a switch cannot guarantee to be able to always connect an idle input to an idle output port, it is said to be internally nonblocking.
It is a mode of operation of packet switched networks in which packets can start only at certain discrete moments in time; each time slot, packets on different channels are aligned.
In this particular mode, packets can arrive at any moment in time, without any alignment.
Despite the fact that both information characteristics and communication technology have greatly evolved since the time Clos’s seminal paper appeared, his ideas on multistage switches still prove to be very useful. In this discussion, the focus has been on their application in optical networking. A range of examples in the field of circuit switching has been outlined, after which the focus has been on Clos-like design in optical packet switching.
Two of the most widespread architectures for OPS have been presented: broadcast and select switches using SOAs, and AWG based switches. The former profits from a Clos-like multistage architecture to reduce the number of SOA gates required and to enlarge the switch size to high port counts. The AWG based design was shown to be prone to internal blocking when the tenability of wavelength converters is limited. A multistage design inspired by Clos networks offers a viable solution for the blocking problem.
As in the “good old days”, multistage approaches are still very useful to reduce costs (that is the number of components used) as well as to circumvent technological limitations.