Photon graphic

Photonic switching: data at the speed of light

The next big revolution in networking is to move data in the form of light from start to finish: but photonic switching means a major step-change for the vendors who build the devices that route data packets around the world's data centres.

Optical switching based on photonics represents a major evolution in data transport for the data centre, initially for large-scale aggregation within networks but eventually coming down to the rack level as silicon photonics becomes widely available and affordable over the next five years.

That will affect all data centres but, for the time being, the direct impact of large-scale optical or photonic switching will be mainly felt by larger enterprises. At the same time though small- and medium-sized enterprises are moving workloads out of in-house data centres to cloud-based services, which themselves are starting to embrace photonics. Effectively, therefore, optical switching is advancing on all fronts, either directly into large enterprises or via the backdoor of the cloud services for SMEs.

The principal attraction of optical switching for these larger-scale data centre cases is economic, but security, availability, capacity and resiliency are significant compelling factors, depending on the particular infrastructure.

"The value proposition for optical circuit-switching really applies first to very large data centres, where the sheer scale of networking generates significant costs," says Daniel Tardent, director of marketing at Calient Technologies, arguably the optical circuit switching vendor most focused on the enterprise market. "Second, the proposition makes sense for those data centres that have moved to single mode fibre in their aggregation networks."

Route to cost savings

The reason optical switching is more attractive for enterprises that already have single-mode fibre, as opposed to the alternative multimode fibre, comes down to economics. Multi-mode fibre has a much larger core diameter of 50-100µm, compared with about 8µm for typical single-mode. This gives it a greater light-gathering capacity, which makes the end-connections simpler with higher tolerances and much cheaper.

The downside is that because the light is not so accurately confined along the line of the fibre, it attenuates more quickly – so that both its capacity and distance range are, by a factor of around 50, a lot less.

For this reason single-mode fibre has been traditionally preferred for long distance telco network trunks, where the additional capacity is vital, and the greater range means that fewer electronic components are needed to regenerate the signal, even if they are more expensive. Multi-mode was preferred for local access or metropolitan connections and data centre networks, where the shorter distances and lower-bandwidth requirements meant that the cost of terminating electronics became prohibitive.

That situation is changing, though, with growing capacity requirements of large data centres so that increasingly single-mode is deployed there too. So far most fibre in data centres is used within traditional packet-switched networks, but migration to optical switching is much more appealing economically when the existing cable is single mode because money is saved as a result of replacing the electronic connecting components with the lower-cost photonics.

"For single-mode fibre a 10Gbps transceiver is in the $150-$200 range, or around $1,500 for a 40Gbps transceiver," Tardent says. "For multimode the 10Gbps transceiver is about $50, and $400 for 40Gbps – so my point is that, if you look at the total cost of a packet-based switch like a Cisco Nexus 7000, it becomes much more expensive when using single-mode transceivers." So while the optical switch is always less expensive, the saving is that much greater when comparing with a packet solution that has single-mode fibre, Tardent says.

There's also the argument that single-mode has the advantage of being more future-proof, particularly where optical switching is being deployed. As bandwidth demand increases it is quite likely many large datacentres will, in future, want to replace multimode with single-mode for the extra performance.

It would, therefore, make sense to do that at the same time as deploying optical switches, since they too should not need replacing for a long time.

Meanwhile, Calient's S Series circuit switch can support any bit-rate that can be carried by the fibres without limit as it deals in wavelengths rather than aggregations of data packets. The throughput is determined solely by the number of wavelengths carried by each fibre, as well as the amount of data encoded onto each wavelength – both of which depend on the current state of the end-point technology.

The same should apply to competing optical switches from the few other vendors in the market such as Polatis, CrossFiber, Alcatel-Lucent, Glimmerglass Cyber Solutions and Ciena Corporation.

Apart from cost-savings at the infrastructure level, optical switching will play a key role in two major developments that are changing the shape of the enterprise data centre: cloud computing and software-defined networking (SDN). Most data centres are embracing cloud computing to some degree, either for the bulk of their computing or, at the very least, for some overspill capacity.

At the same time SDN is emerging as a way of making networking simpler, more flexible and cost effective. It does this through separation, or abstraction, of the underlying switches and routers where decisions over forwarding of packets, or transmission of wavelengths, are made from the higher-level service management.

"Cloud holds the promise of being able to spill-out beyond the walls of the data centre and move from a capex to an opex model where these assets, storage, compute and infrastructure are consumed as needed, more like a utility model," suggests Kevin Drury, senior director of optics products and solutions in Alcatel-Lucent's IP Routing & Transport Division.

"Photonic switching will be an absolutely core technology enabling these cloud-based services, where multiple enterprise data centres come in to handle peak demands that may be time-of-day related, or for seasonal quarterly earnings closing. They will be able to dynamically turn up a wavelength to provide the capacity."

In the case of SDN, optical switching is not itself essential; but the motive for deploying SDN is to enable rapid and automated deployment of network resources at whatever scale is required, which in the case of global cloud services of the kind provided by, say, IBM, Google, Amazon or Dimension Data, involves massive re-allocation of bandwidth, storage and compute power between clients. This can only be done effectively with the aid of optical switching to minimise transmission latency and enable services to be conveniently mapped to wavelengths.

A key requirement to make SDN work is an effective control plane to separate resource management from the underlying network fabric or data plane, which will typically comprise a hybrid between traditional packet-based components such as routers and optical switches.

Is OpenFlow go?

It had been hoped and expected that OpenFlow, developed specifically by the Open Networking Foundation (ONF) for controlling and manipulating routers and switches under SDN, would step-up to the plate as the universal control plane protocol.

It may well do so but, according to Calient's Tardent, it has been a long time coming. "To some extent we've been hoping that the growth in SDN controllers and different variations of OpenFlow would magically create a solution for controlling hybrid fabrics in a data centre," he says, "but now we're actually doing a control plane development ourselves for a controller to monitor the traffic matrix in the network and that can intelligently move flows between packet and circuit switched networks based on traffic or priority. It's really meant to ultimately plug into any SDN architecture the data centre may be deploying."

In the longer run data centres may be looking for standards-based controllers that are not associated with a given make of optical switch, and this is where the OpenFlow communications protocol should enable some degree of interoperability at the control plane level. OpenFlow is, however, only essentially a communications protocol and does not deliver all the integration and interoperability that will be required or be desirable between the optical and traditional IP packet-switched world.

A more sophisticated protocol has evolved to provide this degree of interoperability, called Generalised Multi-Protocol Label Switching (GMPLS). This is an extension of the venerable MPLS, which was developed in the late 1990s to speed-up IP packet transmission through multi-hop router networks by replacing long network addresses with short labels. This avoided complex lookups in routing tables and effectively enabled virtual end-to-end paths to be created through router networks for the duration of a session.

GMPLS, which is being strongly pushed by Cisco, raises the protocol to the control plane of a hybrid network by extending the label-swapping concept beyond routers to different types of networking equipment, but particularly optical switches. It also enables resiliency and network restoration by bridging the IP and optical domains.

This is significant in the light of a recent survey from ICT analyst Ovum which found that enterprises ranked resiliency as one of the three main impediments to adoption of cloud services, along with latency and security. Optical switching reduces latency at the aggregate level, while silicon photonics may in future do so closer to the end points within the data centre.

Latency is becoming an increasing concern for particular industries such as TV broadcasting, which is undergoing radical changes in distribution models from one-to-many transmission over satellite and terrestrial links, to unicast one-to-one over the Internet, where speed and latency assurances are critical.

This rise in OTT (over the top) content services from the likes of Netflix, Amazon and Hulu – as well as broadcasters' portals such as the BBC iPlayer in the UK – is one factor that's changing the shape of global data centre networks because it requires video content to be located or staged closer to consumers to minimise latency. It also requires the rapid recruitment of network resources that optical switching enables.

Optical switching can also score on security, again at the aggregate level, claims Alcatel-Lucent's Drury, because of the ability to protect data at the wavelength level: "Service providers could offer secure wavelengths using strong AES [advanced encryption standard] 256 encryption, so the wavelength itself is inherently secure."

From the perspective of technological innovation rising to the challenges of demands for ever-faster data delivery, it looks as if all roads in the data centre are leading towards optical switching. However, the technology is still too complex to be deployed by smaller data centres that lack sufficient internal resources. It will, though, sweep through the cloud services that many smaller enterprises will be reliant upon. 

Further information

Recent articles

Info Message

Our sites use cookies to support some functionality, and to collect anonymous user data.

Learn more about IET cookies and how to control them

Close