Speed is everything in the data center. Today, to futureproof effectively, providers have to look ahead and assess not only what’s needed now, but what must be delivered in the future.
As a result of exponential demand, this means network design must continue to support massive amounts of traffic and, crucially, allow for server, storage and network capacity to scale. Of course, this must be achieved with as little disruption and reconfiguration as possible; the network infrastructure within the data center must be scalable to support higher server densities, more fiber, and migration to higher speeds.
The discussion around migration to higher line rates is both complex and always evolving. Data center professionals must make a plethora of decisions regarding fiber type, modulation and transmission schemes, connector configurations and, importantly, cost considerations. There are several aspects to carefully consider when determining which migration path is best for any given environment. Here are just four of the issues which must be navigated regarding high-speed data center migration.
Until recently, the accepted Ethernet migration roadmap detailed an expected jump from 10G lanes to 40G. However, since the approval of the IEEE 802.3by standard, the industry has seen a shift to 25G lanes as the next switching technology. The newer 25G lanes are favoured because they offer easy migration to 50G (2x25G) and 100G (4x25G). To a lesser extent, they also offer improved utilization of the switching silicon in network switches. Using a network port at 25G vs 10G provides more capacity for the same capital and operating costs. 25G lanes also enable a logical grouping for support of 100G, 200G, and 400G speeds.
Modulation schemes are also becoming increasingly available. Pulse-amplitude modulation with four amplitude levels (PAM-4) has been proposed for optical links, within the data center and among multiple data center facilities. The PAM-4 uses four distinct pulse amplitudes to transmit data, twice the transmission capacity of the traditional NRZ, at the same signaling rate. However, the downside is that it requires a higher signal-to-noise ratio (SNR), placing stricter requirements on the supporting physical infrastructure. Nevertheless, the PAM-4 remains one of the most promising modulation techniques for 100G and beyond, due to its simplicity and low power consumption.
Various wavelength division multiplexing (WDM) techniques are being developed to increase the number of wavelengths transmitted on each fiber. For more than two decades, WDM has been used to increase data rates on long-haul networks by reducing fiber counts. It has also been used for singlemode Ethernet applications, such as 10GBASE-LR4 and 100GBASE-LR4 which combine four wavelengths on the same fiber using coarse WDM technology. This concept has also been extended to multimode fiber using a technique known as shortwave WDM or SWDM; this utilizes wavelengths from 850nm to 940nm.
The market is gravitating to parallel optics as more demanding applications drive higher data rates. Using laser-optimized multimode fiber (LOMMF), serial optics can cost-effectively support speeds up to 10G. However, using serial transmissions to support 25G or 40G historically required switching to more costly singlemode transceivers. Parallel optics, on the other hand, provide a cost-effective solution for migration to 40G and allow the grouping of 25G lanes to deliver 100G. Future paths for 200/400G are already being established, with ethernet on both singlemode and multimode fiber using a combination of serial and parallel transmission.
The use of MPO connectors is accelerating the switch to parallel optics, with sales of 40/100GbE MPO connectors forecast to increase 15.9 percent annually through 2020. However, the trend to parallel optics may ebb and flow as new technologies are implemented that make better use of duplex pairs. Duplex 100G applications using four 25G lanes are also being driven by cost-effective technologies such as SWDM4 and we can expect 50G PAM-4 lanes to provide 100G over multimode fiber. Both SWDM4 and PAM-4 enable additional cost savings, requiring fewer fibers than an equivalent parallel optic system.
Migration strategies are constantly evolving so evaluating the options and market trajectory is not always easy. However, it’s important to remember that any technical options or emerging solutions must be viewed within the context of your specific enterprise data center environment. CommScope expects that pre-terminated MPO-based fiber solutions will remain the optimal choice for high-performance networks and MPO 12-fiber systems will continue to be used in support of single and parallel applications. Ultimately, you should consider the trajectory of the enterprise, the velocity of change, and the total cost of ownership for the various migration scenarios being considered.
You can learn more about high-speed migration considerations in the data center here.