Huawei's novel Petabit switch
The Chinese equipment maker showcased a prototype optical switch at this year's OFC/NFOEC that can scale to 10 Petabit.
"Although the numbers [400,000 lasers] appear quite staggering, they point to a need for photonic integration"
Reg Wilcox, Huawei
Huawei has demonstrated a concept Petabit Packet Cross Connect (PPXC), a switching platform to meet future metro and data centre requirements. The demonstrator is not expected to be a commercial product before 2017.
Current platforms have switching capacities of several Terabits. Yet Huawei believes a one thousand-fold increase in switching capacity will be needed. Fibre capacity will be filled to 20 and eventually 50 Terabits using higher-order modulation schemes and flexible spectrum. This will add up to a Petabit (one million Gigabits) per site, assuming 200 switched fibres at busy network exchanges.
"We are not saying we will introduce a 10 Petabit product in five years' time, although the technology is capable of that," says Reg Wilcox, vice president of network marketing and product management at Huawei. "We will size it to what we deem the market needs at that time."
Source: Huawei
The PPXC uses optical burst transmission to implement the switching. Such burst transmission uses ultra-fast switching lasers, each set to a particular wavelength in nanoseconds. Like Intune Networks’ Verisma iVX8000 optical packet switching and transport system, each wavelength is assigned to a particular destination port. As OTN traffic or packets arrive, they are assigned a wavelength before being sent to a destination port.
Huawei's switch demonstration linked two Huawei OSN8800 32-slot platforms, each with an Optical Transport Network (OTN) switching capacity of 2.56 Terabit-per-second (Tbps), to either side of the core optical switch, to implement what is known as a three-stage Clos switching matrix.
With each OSN8800, half the slots are for inter-machine trunks to the core optical switch, the middle stage of the Clos switch. "The other half [of the OSN8800] would be dedicated to whatever services you want to have: Gigabit Ethernet, 10 Gigabit Ethernet; whatever traffic you want riding over OTN," says Wilcox.
The core optical switch implements an 80x80 matrix using 80 wavelengths, each operating at 25Gbps. The 80x80 matrix is surrounded by MxM fast optical switches to implement a larger 320x320 matrix that has an 8 Terabit capacity. It is these larger matrices - 'switch planes' - that are stacked to achieve 10 Petabit. The PPXC grooms traffic starting at 1 Gigabit rates and can switch 100Gbps and even higher speed incoming wavelengths in future.
Oclaro provided Huawei with the ultra-fast lasers for the demonstrator. The laser - a digital supermode-distributed Bragg reflector (DS-DBR) - has an electro-optic tuning mechanism, says Robert Blum, director of product marketing for Oclaro's photonic component. Here current is applied to the grating to set the laser's wavelength. The resulting tuning speed is in nanoseconds although Oclaro will not say the exact switching speed specified for the switch.
Each switch plane uses 4x80 or 320, 25Gbps lasers. A 10 Petabit switch requires 400,000 (320x1250) lasers. "Although the numbers appear quite staggering, they point to a need for photonic integration," says Wilcox. Huawei recently acquired photonic integration specialist CIP Technologies.
The demonstration highlighted the PPXC switching OTN traffic but Wilcox stresses that the architecture is cell-based and can support all packet types: "We are flexible in the technology as the world evolves to all-packet.” The design is therefore also suited to large data centres to switch traffic between servers and for linking aggregation routers. "It is applicable in the data centre as a flattened [switch] architecture," says Wilcox.
Huawei claims the Petabit switch will deliver other benefits besides scalability. "Rough estimates comparing this device to OTN switches, MPLS switches and routers yields savings of greater than 60% on power, anywhere from 15-80% on footprint and at least a halving of fibre interconnect," says Wilcox.
Meanwhile Oclaro says Huawei is not the only vendor interested in the technology. "We have seen quite some interest recently in this area [of optical burst transmission]." says Oclaro's Blum. "I wouldn't be surprised if other companies make announcements in this space."
Further reading:
- OFC/ NFOEC 2012 paper: An Optical Burst Switching Fabric of Multi-Granularity for Petabit/s Multi-Chassis Switches and Routers
Packet optical transport: Hollowing the network core
The platform enables a fully-meshed metropolitan network
Intune Networks' CEO, Tim Fritzley (right) and John Dunne, co-founder and CTO with software support for web-based services, claims the Irish start-up.
“What we have designed allows for the sharing of the same fibre switching assets across multiple services in the metro,” says Tim Fritzley, Intune’s CEO.
The company is in talks with several operators about its OPST system, which is being used for a nationwide network in Ireland. The system is also part of an EC seventh framework project that includes Spanish operator Telefónica.
OPST architecture
Intune’s OPST system, dubbed the Verisma iVX8000, uses dense wavelength division multiplexing (DWDM) technology but with a twist. Each wavelength is assigned to a particular destination port, over which the data is transmitted in bursts. The result is an architecture that uses both wavelength-division and time-division multiplexing.
To enable the approach, Intune has developed a control algorithm that can switch and lock a tunable laser’s wavelength “in nanoseconds”. Such rapid laser switching enables wavelength addressing - assigning a dedicated wavelength to each destination port.
As packets arrive at the iVX8000, they are ‘coloured’ and queued before being sent on the required wavelength to their destination. In effect packets are routed at the optical layer, in contrast to traditional systems where traffic is packed onto a lightpath that has a fixed predefined point-to-point optical path.
The packets are sent in bursts based on their class-of-service. Intune uses a proprietary framing scheme for transmission, with Ethernet frames restored at the destination. At the input port, all the packets are queued based on their wavelength and class-of-service. The scheduler, which composes the bursts, picks bits to transmit from the queues based on their class, with the bits sent without having to be aligned with a frame’s boundaries.
“Instead of assigning an electrical address to a fixed wavelength, we are assigning electrical addresses to dynamic wavelengths”
Tim Fritzley, Intune Networks
Intune also uses dynamic bandwidth allocation: any bandwidth unused by the higher classes of service is assigned to lower priority traffic. This achieves over 80 percent utilisation of the Ethernet switching and the fibre, says Fritzley.
“You are responding to the dynamic loading of the traffic as it comes in, on a destination-by-destination, colour-by-colour basis,” says Fritzley “Instead of assigning an electrical address to a fixed wavelength [as with traditional systems], we are assigning electrical addresses to dynamic wavelengths.”
The result is a fully meshed architecture with any transponder able to talk to any other transponder on the network, says Fritzley.
System’s span
The network architecture is arranged as a ring with up to a 300km span. The ring connects up to 16 iVX8000 nodes each comprising four 10 Gigabit-per-second (Gbps) ports and switching hardware. Each port is assigned a particular wavelength, equating to a total switch capacity of 640Gbps.
Intune has an 80-wavelength design even though only 64 are used. Indeed it uses two optical rings in parallel. The two rings run in opposite directions, providing optical protection for each port and effectively doubling overall capacity.
For the client side interfaces, the iVX8000 uses four 10 Gigabit Ethernet ports. Since transmissions are in bursts, multiple ports can transmit data to the same destination port even though they share the same wavelength.
The system’s 300km span is an artificial value set by Intune to guarantee “plug-and-play” performance. If the individual chassis are less than 65km apart and the total ring is 300km or under, Intune guarantees no DWDM engineering is required. “We auto-discover all the optical paths and nodes in the network; we automatically adjust all the amplification and set up the dispersion compensation,” says Fritzley. “This saves thousands of engineer-hours and truck rolls.”
Intune points out that it has engineered a 700km network but claims that for distances beyond 1,000km, point-to-point links connecting regions make more sense.
John Dunne, co-founder and CTO of Intune, claims the metro architecture simplifies networking greatly when connecting the network edge to the IP core. “It is different to what is there today because there are no routeing decisions to be made,” says Dunne. “All of the routes pre-exist, and that is because the tunable lasers contain all the colours of all the ports on the ring.”
As a result, setting up a flow of packets between the edge and core involves using a single interface to the ring. “You don’t have to talk to all the [ring’s] elements, you just talk to the ring,” says Dunne. “The ring is pre-engineered so it knows it’s a ring; it also knows how to guarantee the latency, the bandwidth, the jitter of any flow.”
This is the system’s main merit, says Dunne, the pre-engineered ring hides all the difficulty of building a control layer on top of a dynamic optical and layer-two switching system.
Bringing the web into the network
Intune realised that traditional telecom software would struggle to make best use of its distributed optical packet switch architecture. The company has adopted the representational state transfer (REST) software approach for its architecture instead.
“REST is the heart of web services,” says Fritzley. “The reason we did this is that there are hundreds of thousands of programmers that understand how to program it, so you are not into the arcane telecom languages of SNMP and TL1.” Adopting a 'RESTful' approach, claims Intune, reduces code development by 70 percent.
Moreover, REST by its nature is distributed such that it lends itself to supporting distributed transactions across Intune’s switch. “We have put a mini-http server on every card; we do not centralise control inside a node,” says Fritzley. “Every card peers with all of its peer-functions on the ring.”
In terms of the switch's operation, high-level XML commands are used instead of sending low-level instructions to numerous elements. “For example you ask the ring - set up this flow of packets with this bandwidth, this jitter and this delay,” says Dunne. “The ring replies that it can set this up and it performs the low-level stuff internal to the ring."
Such a capability will ultimately enable a machine to provision bandwidth for services, and enable machine-to-machine communications, says Intune. It will also enable third-party application developers to use the switch for service provisioning. This isn’t possible today because there is a lack of control, says Dunne.
“We have a full suite of XML-based interface commands,” he says. “All [the interface commands] would go to the carrier, the carrier would expose a subset to the Googles, the Googles would expose a subset to their application writers, and the application writers would expose a subset to the consumer.” Were the consumer to send a command to request some bandwidth, the call would be passed through the various layers directly into the switch, all in a controlled manner.
Provisioning of bandwidth in such an automated fashion is possible because Intune’s underlying network is bounded and predictable, says Dunne, with the optical path pre-engineered to work with the data path.
Meanwhile until XML becomes more commonplace, Intune uses a code translator that converts the XML code to SNMP or TL1 to interface to existing systems.
“The ring is pre-engineered so it knows it’s a ring; it also knows how to guarantee the latency, the bandwidth, the jitter of any flow”
John Dunne, Intune Networks
Applications
The iVX8000 is being targetted at applications such as cloud computing services and the moving of virtualised environments between data centres. But the real target is using the platform to support multiple services – 3G and 4G wireless backhaul, on-demand IP TV as well as cloud. “No-one can do traffic planning anymore around such services,” says Fritzley.
The platform addresses what one large European operator calls ‘hollowing the core’. The operator wants to simplify its metro network by moving such networking elements as broadband remote access servers (BRASs) to the network edge. These will be connected using a simpler layer-two network that lessens the use of large, expensive IP core routers.”All the IP processing is on the edge and you go edge-to-edge on a flat layer two,” says Fritzley.
Market developments
Intune is using its system to enable the Exemplar network in Ireland. Backed by the Irish Government, the company’s systems will be used to build a nationwide network. The first phase involves a lab for application development and testing. So far 40 multi-nationals have signed up to use the network. Starting next year, a ring network will be up and running around Dublin to be followed with a nationwide roll-out in 2013.
The Irish start-up is also part of an EC Seventh Framework research project called MAINS. The project, which started in January, involves Telefónica which is using the iVX8000 to move virtualised resources between data centres depending on user demand and latency requirements. The project uses XML commands to call for bandwidth from the networking layer.
Meanwhile, Intune says that it is “deeply engaged” with four to five of the largest operators in North America and Europe.