QSFP28 MicroMux expands 10 & 40 Gig faceplate capacity
- ADVA Optical Networking's MicroMux aggregates lower rate 10 and 40 gigabit client signals in a pluggable QSFP28 module
- ADVA is also claiming an industry first in implementing the Open Optical Line System concept that is backed by Microsoft
The need for terabits of capacity to link Internet content providers’ mega-scale data centres has given rise to a new class of optical transport platform, known as data centre interconnect.
Source: ADVA Optical Networking
Such platforms are designed to be power efficient, compact and support a variety of client-side signal rates spanning 10, 40 and 100 gigabit. But this poses a challenge for design engineers as the front panel of such platforms can only fit so many lower-rate client-side signals. This can lead to the aggregate data fed to the platform falling short of its full line-side transport capability.
ADVA Optical Networking has tackled the problem by developing the MicroMux, a multiplexer placed within a QSFP28 module. The MicroMux module plugs into the front panel of the CloudConnect, ADVA’s data centre interconnect platform, and funnels either 10, 10-gigabit ports or two, 40-gigabit ports into a front panel’s 100-gigabit port.
"The MicroMux allows you to support legacy client rates without impacting the panel density of the product," says Jim Theodoras, vice president of global business development at ADVA Optical Networking.
Using the MicroMux, lower-speed client interfaces can be added to a higher-speed product without stranding line-side bandwidth. An alternative approach to avoid wasting capacity is to install a lower-speed platform, says Theodoras, but then you can't scale.
ADVA Optical Networking offers four MicroMux pluggables for its CloudConnect data centre interconnect platform: short-reach and long-reach 10-by-10 gigabit QSFP28s, and short-reach and intermediate-reach 2-by-40 gigabit QSFP28 modules.
The MicroMux features an MPO connector. For the 10-gigabit products, the MPO connector supports 20 fibres, while for the 40-gigabit products, it is four fibres. At the other end of the QSFP28, that plugs into the platform, sits a CAUI-4 4x25-gigabit electrical interface (see diagram above).
“The key thing is the CAUI-4 interface; this is what makes it all work," says Theodoras.
Inside the MicroMux, signals are converted between the optical and electrical domains while a gearbox IC translates between 10- or 40-gigabit signals and the CAUI-4 format.
Theodoras stresses that the 10-gigabit inputs are not the old 100 Gigabit Ethernet 10x10 MSA but independent 10 Gigabit Ethernet streams. "They can come from different routers, different ports and different timing domains," he says. "It is no different than if you had 10, 10 Gigabit Ethernet ports on the front face plate."
Using the pluggables, a 5-terabit CloudConnect configuration can support up to 520, 10 Gigabit Ethernet ports, according to ADVA Optical Networking.
The first products will be shipped in the third quarter to preferred customers that help in its development while the products will be generally available at the year-end.
ADVA Optical Networking unveiled the MicroMux at OFC 2016, held in Anaheim, California in March. ADVA also used the show to detail its Open Optical Line System demonstration with switch vendor, Arista Networks.
Two years after Microsoft first talked about the [Open Optical Line System] concept at OFC, here we are today fully supporting it
Open Optical Line System
The Open Optical Line System is a concept being promoted by the Internet content providers to afford them greater control of their optical networking requirements.
Data centre players typically update their servers and top-of-rack switches every three years yet the optical transport functions such as the amplifiers, multiplexers and ROADMs have an upgrade cycle closer to 15 years.
“When the transponding function is stuck in with something that is replaced every 15 years and they want to replace it every three years, there is a mismatch,” says Theodoras.
Data centre interconnect line cards can be replaced more frequently with newer cards while retaining the chassis. And the CloudConnect product is also designed such that its optical line shelf can take external wavelengths from other products by supporting the Open Optical Line System. This adds flexibility and is done in a way that matches the work practices of the data centre players.
“The key part of the Open Optical Line System is the software,” says Theodoras. “The software lets that optical line shelf be its own separate node; an individual network element.”
The data centre operator can then manage the standalone CloudConnect Open Optical Line System product. Such a product can take coloured wavelength inputs and even provide feedback with the source platform, so that the wavelength is tuned to the correct channel. “It’s an orchestration and a management level thing,” says Theodoras.
Arista recently added a coherent line card to its 7500 spine switch family.
The card supports six CFP2-ACOs that have a reach of up to 2,000km, sufficient for most data centre interconnect applications, says Theodoras. The 7500 also supports the layer-two MACsec security protocol. However, it does not support flexible modulation formats. The CloudConnect does, supporting 100-, 150- and 200-gigabit formats. CloudConnect also has a 3,000km reach.
Source: ADVA Optical Networking
In the Open Optical Line System demonstration, ADVA Optical Networking squeezed the Arista 100-gigabit wavelength into a narrower 37.5GHz channel, sandwiched between two 100 gigabit wavelengths from legacy equipment and two 200 gigabit (PM-16QAM) wavelengths from the CloudConnect Quadplex card. All five wavelengths were sent over a 2,000km link.
Implementing the Open Optical Line System expands a data centre manager’s options. A coherent card can be added to the Arista 7500 and wavelengths sent directly using the CFP2-ACOs, or wavelengths can be sent over more demanding links, or ones that requires greater spectral efficiency, by using the CloudConnect. The 7500 chassis could also be used solely for switching and its traffic routed to the CloudConnect platform for off-site transmission.
Spectral efficiency is important for the large-scale data centre players. “The data centre interconnect guys are fibre-poor; they typically only have a single fibre pair going around the country and that is their network,” says Theodoras.
The joint demo shows that the Open Optical Line System concept works, he says: “Two years after Microsoft first talked about the concept at OFC, here we are today fully supporting it.”
Ethernet access switch chip boosts service support
The Serval-2 architecture. Source: Vitesse
Vitesse Semiconductor has detailed its latest Carrier Ethernet access switch for mobile backhaul, cloud and enterprise services.
The Serval-2 chip broadens Vitesse's access switch offerings, adding 10 Gigabit Ethernet (GbE) ports while near-tripling the switching capacity to 32 Gigabit; the Serval-2 has 2x10 GbE and 12 1GbE ports.
The device features Vitesse's service aware architecture (ViSAA) that supports Carrier Ethernet 2.0 (CE 2.0). "We have built a hardware layer into the Ethernet itself which understands and can provision services," says Uday Mudoi, product marketing director at Vitesse.
CE 2.0, developed by the Metro Ethernet Forum (MEF), is designed to address evolving service requirements. First equipment supporting the technology was certified in January 2013. What CE 2.0 does not do is detail how services are implemented, says Mudoi. Such implementations are the work of the ITU, IETF and IEEE standard bodies with protocols such as Muti-Protocol Label Switching (MPLS)/ MPLS-Transport Profile (MPLS-TP) and provider bridging (Q-in-Q). "There is a full set of Carrier Ethernet networking protocols which comes on top of CE 2.0," says Mudoi.
Serval-2 switch
The Serval-2 switch features include 256 Ethernet virtual connections, hierarchical quality of service (QoS), provider bridging, and MPLS/ MPLS-TP.
An Ethernet Virtual Connection (EVC) is a logical representation of an Ethernet service, says Vitesse, a connection that an enterprise, data center or cell site uses to send traffic over the WAN.
Multiple EVCs can run on the same physical interface and can be point-to-point, point-to-multipoint, or multipoint-to-multipoint. Each EVC can have a bandwidth profile that specifies the committed information rate (CIR) and excess information rate (EIR) of the traffic transmitted to, or received from, the Ethernet service provider’s network.
The EVC also supports one or more classes of service and measurable QoS performance metrics. Such metrics include frame delay - latency - and frame loss to meet a particular application performance requirements.
The Serval-2 supports 256x8 class of service (CoS) EVCs, equivalent to over 4,000 bi-directional Ethernet services, says Mudoi.
The Serval-2 also supports per-EVC hierarchical queuing. It allows for 256 bi-directional EVCs with policing, statistics, and QoS guarantees for each CoS and EVC. Hierarchical QoS also enables a mix of any strict or byte-accurate weighting within the EVC, and supports the MEF's dual leaky bucket (DLB) algorithm that shapes traffic per-EVC and per-port.
"Service providers guarantee QoS to subscribers for the services that they buy," says Mudoi. "If each subscriber's traffic - even different applications per-subscriber - is treated using separate queues, then one subscriber's behavior does not impact the QoS of another." Supporting thousands of queues allows service providers to offer thousands of services, each with its own QoS.
Q-in-Q, defined in IEEE 802.1ad, allows for multiple VLAN headers - tags - to be inserted into a frame, says Mudoi, enabling service provider tags and customer tags.
Meanwhile, MPLS/ MPLS-TP direct data from one network node to the next based on shortest path labels rather than on long network addresses, thereby avoiding complex routing table look-ups. The labels identify virtual links between distant nodes rather than endpoints.
MPLS can encapsulate packets of various network protocols. Serval-2's MPLS-TP supports Label Edge Router (LER) with Ethernet pseudo-wires, Label Switch Router (LSR), and H-VPLS edge functions.
Q-in-Q in considered a basic networking function for enterprise and carrier networks, says Mudoi, while MPLS-TP is a more complex protocol.
Serval-2 also supports service activation and Vitesse's implementation of the IEEE 1588v2 timing standard, dubbed VeriTime.
"Before you provision a service, you need to run a test to make sure that once your service is provisioned, the user gets the required service level agreement," says Mudoi. Serval-2 supports the latest ITU-T Y.1564 service activation standard.
IEEE 1588v2 establishes accurate timing across a packet-based network and is used for such applications as mobile. The Serval-2 also benefits from Intellisec, Vitesse's MACsec Layer 2 security standard implementation (see Vitesse's Intellisec ).
"Both [Vitesse's VeriTime IEEE 1588v2 and Intellisec technologies] highly complement what we are doing in ViSAA," says Mudoi.
Availability
Serval-2 samples will be available in the third quarter of 2013. Vitesse expects it will take six months for system qualification such that Ethernet access devices using the chip and carrying live traffic are expected in the first half of 2014.