Infinera speeds up network restoration

• Claimed to be the only hardware implementation of the Shared Mesh Protection protocol
• Provides network-wide protection against multiple network failures
• The chip is already within the DTN-X system; protocol will be activated this year

Pravin Mahajan, Infinera

Infinera has developed a chip to speed up network restoration following faults.

The chip implements the Shared Mesh Protection (SMP) protocol being developed by the International Telecommunication Union (ITU) and the Internet Engineering Task Force (IETF) and Infinera believes it is the only vendor with hardware acceleration of the protocol.

The SMP standard is still being worked on and will be completed this year. Infinera demonstrated its hardware SMP implementation at OFC/NFOEC 2013 and will activate the scheme in operators' networks using a platform software upgrade this year.

The chip, dubbed Fast Shared Mesh Protection (FastSMP), is sprinkled across cards within Infinera's DTN-X platform and will be linked to other FastSMP ICs across the network. The FastSMP chips exchange signalling information and use internal look-up tables with pre-calculated routing data to determine the required protection action when one or more network failures occur.

Network faults

The causes of network faults range from fibre cuts from construction work to natural disasters such as Hurricane Sandy and the Asia Pacific tsunami. Level 3 Communications cited in 2011 that squirrels were the second most common cause of fibre cuts after construction work. The squirrels, chewing through fibre, accounted for 17 percent of all cuts. Meanwhile, one Indian service provider says it experiences 100 fibre cuts nationwide each day, according to Infinera.

Operators are also having to share their network maps with enterprises that want to assess the risk based on geography before choosing a service provider. "End customers no longer necessarily trust the service level agreements they have with operators," says Pravin Mahajan, director, corporate marketing and messaging at Infinera.  In riskier regions, for example those prone to earthquakes, enterprises may choose two operators. "A form of 1+1 protection,” says Mahajan.

Operators want resilient networks that adapt to faults quickly, ideally within 50ms, without adding extra cost.

Traditional resiliency schemes include SONET/SDH’s 1+1 protection. This meets the sub-50ms requirement but addresses single faults only and requires dedicated back-up for each circuit. At the IP/MPLS (Internet Protocol/ Multiprotocol Label Switching) layer, the MPLS Fast Re-Route scheme caters for multiple failures and is sub-50ms. But it only addresses local faults, not the full network. And being packet-based - at a higher layer of the network - the scheme is costlier to implement.

"End customers no longer necessarily trust the service level agreements they have with operators"

Infinera's protection scheme uses its digital optical networking approach based on its photonic integrated circuits (PICs) coupled with Optical Transport Networking (OTN). OTN resides between the packet and optical layers, and using a mesh network topology, it can handle multiple failures. By sharing bandwidth at the transport layer, the approach is cheaper than at the packet layer.  But being software-based, restoration takes seconds.

Infinera has speeded up the scheme by implementing SMP with its chip such that it meets the 50ms goal.


FastSMP chip

Infinera plans for multiple failures using the Generalized Multiprotocol Label Switching (GMPLS) control plane. “The same intelligence is now implemented in hardware [using the FastSMP processor],” says Mahajan.

The chip is on each 500 Gigabit-per-second (Gbps) line card, within the platform's OTN switch fabric, the client side and as part of the controller. The FastSMP, described as a co-processor to the CPU, hosts look-up tables with rules as to what should happen with each failure. The chips, located in the platform and across the network, then adjust to the back-up plan for each service failure.

Infinera says that the protection is at the service level not at the link level. "It does this at ODU [OTN's optical data unit] granularity," says Mahajan; each circuit can hold different sized services, 2.5 Gigabit-per-second (Gbps)  or 10Gbps for example, all carried within a 100Gbps light path. "By defining failure scenarios on a per-service basis, you now need to put all these entries in hardware," says Mahajan.

To program the chip, network failures are simulated using Infinera's network planning tool to determine the required back-up schemes. These can be chosen based on shortest path or lowest latency, for example.

The GMPLS control plane protocol determines the rules as to how the network should be adapted and these are written on-chip. When a failure occurs, the chip detects the failure and performs the required actions.

The FastSMP chip is already on all the DTN-X line cards Infinera has shipped and will be enabled using software upgrade.

The GMPLS control plane recomputes backup paths after a failure has occurred. Typically no action is required but if several failures occur, the new GMPLS backup paths will be distributed to update the FastSMPs' tables. "Only on the third or fourth failure typically will a new backup plan be needed," says Mahajan.

In effect, the more meshed the network topology, the greater the number of failures that can be tolerated. "When you have three or four failures, you need to have new computation at the GMPLS control plane and then it can repopulate the backups for failures 3, 4, and 5," he says.


Instant bandwidth and FastSMP

Infinera is able to turn up bandwidth in real-time using its 500Gbps super-channel PIC. "We slice up the 500 Gig capacity available per line card into 100 Gig chunks," says Mahajan.  

This feature, combined with FastSMP, aids operators dealing with failures once traffic is rerouted. The next backup route, if it is close to its full capacity, can have an extra 100 Gigabit of capacity added in case the link is called into use.

A study based on an example 80-node network by ACG Research estimates that the Shared Mesh Protection scheme uses 30 percent less line-side ports compared to an equivalent network implementing the 1+1 protection scheme.