Fiber Mountain’s Glass Core has been recognized as an innovation in the data center space multiple times since its launch in 2014. The Glass Core is based on the company’s Optical Path Exchange (OPX) product line, which provides a very low latency 5-nanosecond transit time. As we discuss applications for the OPX, many times we look at the merits of Optical-Electrical-Optical (OEO) and Optical-Optical-Optical (OOO) technology.
The debate about which technology is better has gone on for more than a decade, and there are good reasons to deploy either solution. However, with advancements in electrical cross-point switching in recent times, there are significant advantages that the OEO architecture offers that were not available a decade ago.
One of the common beliefs over time has been that OEO technology introduces more latency than its OOO counterpart, but recent advancements in electrical cross point switching have narrowed this gap significantly, to the point that latency alone may itself be insignificant as a reason to select one technology over the other. Although we can find merits of one or the other technology based on individual use cases and applications, latency should no longer be considered a factor. The evidence to support this can be found in one of the most latency sensitive business environments in our industry: The high frequency trading (HFT) environment.
In HFT, nanoseconds can be the difference between gaining and losing millions of dollars. Many of these environments are using OEO technology to receive data from traders which is then fed to their super-fast compute platforms. An advantage that OEO platforms can provide is the point to multipoint transmission, which enables many applications requiring replication of data. These include tap and monitoring, video distribution, data broadcast, and low latency data feeds for HFT and streaming applications. In the particular case of HFT, the fact that OEO is the choice for this most latency sensitive application proves that the difference in latency of OEO verses OOO is not a limitation and we need to look at all benefits when selecting a solution.
In full disclosure, the Fiber Mountain OPX family of products is based on OEO technology which may reveal a bias for this post. However, the evidence provided above, as well as ongoing discussions with our customers, is proving to us every day that OEO is gaining an unprecedented foothold in the data center industry, and will be the key component in layer 1 switching solutions which we will see more of in 2016.
High level overview of OEO verses OOO
- OEO receives from the network, an optical “O” signal, converts it to electrical “E”, then switches to a different port, then converts it back to optical “O” and returns it to the network
- OOO receives from the network, an optical “O” signal, switches it to a different port in the optical “O” domain, then returns it back to the network as an optical “O” signal
Let’s delve into an important and quick overview of how these technologies work.
The following simple diagram demonstrates how an OEO switch is constructed:
As we see here, a signal is received by a component at “B”, which represents an optical to electrical converter. Once the conversion takes place, the electrical signal is passed on to the electrical switching function “A”, which switches the signal to one or more ports. The signal is converted back from electrical to optical at “B” once again. From here, the signal is then transmitted on to the network.
The function of “B” above—which is described here as optical/electrical converter—can be implemented in several ways. For instance, one implementation of this is a transceiver, such as an SFP+. In the case of Fiber Mountain’s OPX, we have implemented this function using on-board optics which eliminates the need for external SFP+ and QSFP+ transceivers. This and the use of 24 fiber MPO connectors on the front panel enables the OPX to leverage 160 ports within a one rack unit of space, making it the industry’s most dense optical switch.
The following simple diagram demonstrates how an OOO switch is constructed:
In typical implementations signal is received from the network, and is reflected from the mirror C and on to mirror D—where mirror C and mirror D are movable and can be adjusted to send the received signal from one port to another specified port. The function of such mirrors is implemented using 3D MEMS (micro-electro-mechanical system) technology which requires tuning, and high-precision alignment. It is important to align the signal carefully and although the mirrors can be adjusted, there needs to be a method to determine whether or not the alignment is optimal. A small percentage of the signal is split out from the input at E and passed on to the feedback comparison block G, while similarly a small percentage of the signal is split out from the output at F and passed on to the comparison block G. The system continuously compares the input and output signal power and adjusts the mirrors for optimal output. This method is repeated for every single optical signal that passes through the OOO switch, and many times is the driving factor in high per port costs of the solution. There are other implementations of OOO switches, some of them using mechanical techniques, which reduce reliability.
Comparing a few characteristics of OEO and OOONow let us compare a few characteristics of these technologies and discuss how we can take advantage of them in our networks. The first characteristic we look at is the optical loss through these devices.
An OOO switch can have optical loss through the MEMs array interface, and additional loss when the system has to bleed off some of the optical signal both at ingress and egress in order to measure the optical power of the signal and use this information to adjust the mirrors for proper alignment of the input fiber to the output fiber.
OEO is not faced with such optical loss, as it regenerates the signal. This signal regeneration is an advantage for data center networks as the signal’s reach is increased. The operational distance between two devices that are communicating through the OEO switch is double as compared to communicating through an OOO. We show this through a 10Gbps interface in the diagrams below.
Another characteristic that is desirable in such systems is the ability to multicast traffic from one port to multiple ports.
The unique OEO design of the OPX allows for simultaneous multicast to a number of destinations. The OPX system receives the signal, duplicates it, and sends an exact copy to as many as 160 other optical ports on the same 1RU device, at full signal strength, meeting the industry standard for optical signal output on each port.
While this multicast functionality is standard in OEO design, this same functionality would have to be built out separately using splitters for OOO. The OOO cross-connect would need to split the signal in order to multicast it. Since the signal has to be split using a tap, which is an additional split of the optical path (first split, however small, was for alignment), it is limited to a single copy and that too at a reduced power level as compared to the full strength of the IEEE 10GBASE-SR specifications that the OEO is capable of delivering.
About Fiber Mountain’s OPX
- OPX uses 24-fiber MPO connectors: A single 24-fiber MPO connector can provide twelve pairs of fiber, hence twelve 10Gbps bidirectional connections. The OPX can support individual simplex or duplex connections of 10Gbps, 40Gbps or 100Gbps. Combinations of different connection rates in the same MPO connector are supported.
- The OPX is appropriate for stand-alone configurations: It comes with a built-in web GUI for point and click and remote configuration and operation. It can be controlled from any SDN controller.
- OPX implemented as a networked device: Along with Fiber Mountain’s orchestration system, the OPX can be implemented as a networked device, where a number of OPX systems can be configured, provisioned and managed from a central SDN controller. Fiber Mountain’s orchestration system includes the unique capability of dynamically discovering the entire network, including the physical layer.
For more information on the OPX system, and how it leverages the Glass Core, click here.