So, in the drive to wring even the tiniest inkling of latency from your trading network, you’ve painstakingly identified the shortest routes between trading venues, information feeds and co-location facilities. Cognizant that eliminating a mile of physical fiber on these routes translates into a 16-microsecond savings in roundtrip latency (and that the difference between winning and losing deals in contemporary electronic trading is measured in much smaller increments), you’ve mapped routes that bypass as many detours under roads and up and down manholes as possible.
The next step is to dive into the guts of the gear enabling optical transport along the fiber network. Amplification, color conversion, dispersion compensation and regeneration are the main optical-networking functions commonly performed by hardware transmitting and receiving light. But the ways that these processes are carried out differs significantly among equipment providers. Managers of financial networks must investigate the impact of each, end-to-end across their infrastructures, in order to ensure that latency budgets are minimized.
- Amplification—As trading venues are further and further apart (for example London to Frankfurt or New York to Chicago), optical signals need a boost to offset the weakening that occurs across an optical span. It is frequently Erbium Doped Fiber Amplifiers (EDFAs) that are utilized to boost signals once transmission distances between trading venues exceeds 30 miles. If you have several of these amplifiers across a path, it can add significant time to transmission—latency in the microseconds in the case of some common architectures, such as high-gain, dual-stage EDFAs. Some vendors offer latency-optimized EDFAs that can cut amplification time in half. Perhaps the most exciting innovation in low-latency transmission are the significant strides in the advancement and application of RAMAN amplification technology in trading networks. One equipment provider, ADVA Optical Networking, has released a counter and co-propagating RAMAN amplifier that removes need for EDFAs in transport networks and has reduced amplification latency by a factor of 16 in some networks.
- Multiplexing—For multiple reasons, it is often desirable to deploy multiple connections among trading venues. This can provide dedicated paths for various trading strategies, and it can allow market data and order entry to be carried separately. Multiplexing achieves these goals by allowing multiple separated traffic signals to be carried simultaneously across the same physical fiber-optic strand. There are major latency differences among multiplexing techniques. One example is combining multiple lower-speed traffic signals, such as five 1Gbit/s signals aggregated into a single 10Gbit/s transport link, using time division multiplexing (TDM). The processing required in TDM adds several microseconds of latency. Another method is wavelength division multiplexing (WDM), in which each signal is assigned a specific color or wavelength of light, and then the various colors are combined and transported. WDM is much faster, measured in 10s of nanoseconds (a nanosecond is 1/1000th of a microsecond); even still, however, there are large differences among various providers’ technology. Some providers, for example, still utilize thin film filters, which are not only slow but also have a different latency for each color.
- Color conversion—When using a WDM network, the signal must be “transponded”—or, converted to a specific color of light—for delivery of multiple colors/signals across glass fiber. Special low-latency transponders have emerged, handling this function in the single-digit nanoseconds. Standard “colorized” optics, on the other hand, will result in transponder latency of several microseconds.
- Dispersion compensation—“Chromatic dispersion” is one of the ways that signals degrade as they travel across optical links, particularly at high speed such as 10Gbit/s. Some optical networks successfully rely on large amounts of dispersion compensating fiber (DCF) to offset the problem, but this is an ineffective strategy for electronic-trading networks. As you can imagine, installing kilometers of DCF renders the physical distance that signals must travel longer and, therefore, the transmission time slower. Trading networks require methods such as fiber Bragg gratings (FBGs) that have been optimized for low-latency applications.
- Regeneration—Eventually, you cannot simply keep amplifying a signal across a connection; you must regenerate. Here again, conventional techniques do not suffice in electronic trading. Low-latency approaches to regeneration might yield latency measured in mere nanoseconds; whereas, methods that are commonly utilized in other areas of networking introduce hundreds of microseconds of delay.
In the next installment, we’ll look at the characteristics of the type of service that a financial firm might adopt in building an infrastructure for low-latency trading. Managed services, fractured or “dim” fiber and private fiber approaches each have valuable attributes, and a single firm might employ multiple services to achieve the range of its business goals.
Read more on low-latency optical transport here.