By JOSEF BERGER
Quad small form factor pluggable plus (QSFP+) transceiver modules using multimode optics have gained increasing traction in very short-reach InfiniBand applications. Building on this success, some transceiver manufacturers have launched development of a new generation of 40-Gigabit Ethernet (40GbE) long-reach singlemode optics in QSFP+ packages.
|Figure 1. A physical comparison of a CFP module (left) versus a QSFP+ 40GBase-LR4 module (right) shows the increased density achievable through optical integration.|
Breakthroughs in power consumption reduction and integration promise a new generation of networking equipment for the storage, data center, and enterprise networking markets where higher densities and lower power have been long awaited–not only for multimode, but also for singlemode fiber. With QSFP+ emerging as the standard for 40GbE, adapting the technology to longer-reach singlemode applications can address many applications beyond 100 m.
To InfiniBand and beyond!
System houses initially developed the QSFP module concept to improve densities in SONET/SDH applications. But the module format quickly gained the favor of vendors for short-reach multimode fiber applications, such as InfiniBand, Gigabit Ethernet, and Fibre Channel. These companies formed a QSFP multisource agreement (MSA) that subsequently pushed for T11 specification SFF-8436, which defined QSFP for higher speeds in the form of the QSFP+.
SFF-8436 modules have proven popular for standards-based 10-Gbps (Single Data Rate) and 20-Gbps (Dual Data Rate) InfiniBand products. One vendor has already introduced a 40-Gbps (Quad Data Rate) InfiniBand switch using QSFP+ transceiver ports.
Meanwhile, the IEEE's High Speed Study Group formed in 2006 to investigate new standards for high-speed Ethernet, initially 100GbE. By mid-2007, interest in 40GbE resurfaced due to favorable cost, size, and power metrics compared to initial 100GbE implementations. Server vendor architectures are not expected to drive more than 40 Gbps of throughput for several years. The IEEE thus expanded its scope to include both 40GbE and 100GbE rates.
|Figure 2. Data center network evolution with the introduction of singlemode QSFP+.|
In response, the module community first investigated the CFP MSA standard package for both 40GbE and 100GbE. While the CFP format proved workable for 100GbE in terms of faceplate density, the same faceplate size is less economical for a 40GbE package. It became clear that moving to 40GbE QSFP+ would significantly improve port density since QSFP+ is less than one-quarter the width of a CFP module (Figure 1).
The move to 40GbE
The progress toward QSFP+ since the 2009 release of the IEEE 802.3ba 40GbE standard has been remarkable compared to the seven years needed for 10GbE transceiver modules to evolve from standards definition to the SFP+ form factor. Yet QSFP+ modules deployed today are only for short-reach multimode applications. That's about to change. In the latest revision, QSFP+ members added singlemode 40GbE applications to the specification, and development work is well underway.
The path is clear on the electrical side of 40GbE QSFP+ development. The SFP+ groundwork is completed and is being applied directly to QSFP+ implementations. Four 10GbE electrical lanes will be used to reach 40GbE. The optical side still requires integration to enable the final form factor. Once this integration takes place, developers will no longer need to continually replace line cards to accommodate new transceiver form factors. The long-term promise of QSFP+ means the 40-Gbps capable switches will be complete.
Singlemode QSFP+ applications
The largest growth area for QSFP+ continues to be in data centers where bandwidth demand continues to increase. In the past few years, server uplinks have begun the migration from 1GbE to 10GbE. This migration is projected to continue for several years.
The architecture within the data center begins with top-of-rack switches that combine several server downlink SFP+ connections. Most uplinks from the switches to the cluster switch are still multimode 10GbE connections; yet, there has been some migration to 40GbE, mainly using QSFP+ modules.
Depending on the size of the data center, connections from the top-of-rack switches to the cluster switch can be in the range of 100 m. As data centers expand to include several floors or several buildings, the expected increased distance will require singlemode fiber (Figure 2). A singlemode QSFP+ 40GbE approach would support longer distances and offer compatibility with existing multimode QSFP+ line cards.
A singlemode approach provides advantages beyond greater reach. Multimode transceivers use MPO connectors to accommodate the 12, 24, or more fibers required to support the necessary data rates. The ribbon cables and connectors are physically larger and more cumbersome than single-fiber cables and LC connectors used in singlemode applications. The use of ribbon cable also makes it more difficult to leverage a higher density switch.
Cost should also be considered. Multimode transceivers have traditionally been perceived as less expensive than singlemode transceivers. Yet, a fair comparison must include the associated fiber cost. A multimode ribbon cable and the necessary MPO connectors are more expensive than one singlemode cable and a pair of LC connectors.
Singlemode links offer better overall performance, longer reach, support for higher speeds, and easier migration. Multimode transceivers offer power consumption savings due to the lower current requirements of VCSEL versus distributed feedback (DFB) lasers. All factors considered, singlemode technology will become increasingly attractive compared with multimode approaches, even for relatively short-reach applications.
Inside the QSFP+ transceiver
Optical component suppliers have been working to integrate technology that will enable singlemode 40GbE QSFP+ transceivers. Once achieved, QSFP+ will offer significant advantages over the CFP design. The QSFP+ will combine four transmitter optical subassemblies (TOSAs) into a single integrated component. Planar lightwave circuit (PLC) or thin film filter structures will integrate the CFP's discrete multiplexing and demultiplexing functions. CDR functionality will move outside the transceiver, eliminating CDR from the module. To the extent possible, laser drivers will use arrays on the transmission side (and potentially on the receive side).
The transition from CFP to QSFP+ not only requires significant optical integration, but also power consumption reduction of the optical and electrical components.
Table 1 offers a quick comparison of the 40GBase-SR4 and 40GBase-LR4 standards and the major components necessary for the modules that will support them. The table also includes information on 40GBase-ER4, a design that enables 40-km distances on single-mode fiber. This transceiver is conceptual and not yet defined by standards; vendors are interested in standardizing the device.
The technology for developing the ER4 involves using an avalanche photo-diode (APD)-based receiver in place of a PIN receiver with the same transmitter structure as the LR4. The R&D investment for ER4 is minimal while the longer reach is significant in the enterprise network. The challenge is providing the power required by the APD receiver. The ER4 provides another application for QSFP+ in the near future.
Singlemode QSFP+ is the future
The future is now, as 40GbE transceivers in a QSFP+ form factor are poised to address both multimode and singlemode applications in a variety of market spaces. The result will enable designers of next-generation systems to develop high-density switches with greater application flexibility.
JOSEF BERGER is the director of product marketing at Opnext, Inc.