Latency refers to the time it takes a packet to transit a network link twice, from you to the server and then back from the server to you. It is also called delay, or Round Trip Time (RTT), or Ping. Latency is affected by link speed, switch transit speed, switch congestion, server processing time, packet size, the distance the packet must traverse, and the medium through which it must traverse. Each adds to the time. Which dominates depends greatly upon conditions. Network latency is expressed in milliseconds (ms), one thousandth of a second. Latencies in today’s U.S. networks range from 15 to 75 ms; they may range on any given line by tens of milliseconds, the range caused by traffic at switches and link congestion for highly shared links such as CATV networks.
Light and electromagnetic signals only travel at the “speed of light” (186,282 miles per second) in a vacuum. Light down a fiber optics transmission line has to contend with impurities and light bouncing along walls of the fiber that slow things down to around 70% of light speed (special fiber optics with air in the middle get well above 90%, but they are very expensive and restricted now to undersea cabling). Electromagnetic signals flowing down a copper line can be as slow as 60% of light speed. Except for signals going to a satellite, all terrestrial signals also encounter amplifiers (repeaters) and switches in the path. The measured time through the transatlantic fiber optic cables from the United States to Britain is about 60 milliseconds (one thousandth of a second), which cannot be improved upon (the cables include repeaters). Transcontinental latencies are similar—63 ms from San Francisco to New York, say. Of this figure 37 ms come from the length of the fiber optics wiring, the rest from switches and repeaters.
Latency has always been a problem, but one hidden from common view. A web page is built through a series of calls and responses, calls from the computer, responses from the server. There is a premium on getting the first byte there fast so a page build can start, and an equal premium on making the whole page appear as quickly as possible. Latency extends the time to build at every call and response. Web page designers do everything in their power to limit the number of calls during the page build. While the Cloud in its primary application has centralized computing and storage, the “Clouds” offered by Google, Apple, and Amazon depend upon Content Delivery Networks (CDN) that locate data centers at the edge of a global network, all over the earth, which centers contain duplicate files for all content of interest, precisely to get close to a customer location so delay, or latency, is minimized.
While IP packets can be as large as 65,535 bytes, the practical upper limit (and now average large Internet packet size) is the 1500 byte Maximum Transmission Unit of Ethernet, which carries IP traffic over home and business routers and local area networks. The full maximum of 65K adds 20 ms of delay at 25 mbps transmission rate, but only 0.5 ms at 1 gigabit. However, at the lower common rate of 1500 bytes, delay added by 25 mbps link speed is less than 0.5 ms. What is hard to compute as a commonplace is the effect of total packets required to send an information unit before a response can be generated. Without compression a typical advertising image of 728 x 90 pixels with 16 bit color requires 260 packets; if compressed best chances are 10:1 ratio, or 26 packets, which takes more than 12 ms at 25 mbps to transfer. However, at 1 gigabit, the transfer rate is 0.3 ms.
We offer these numbers because a variety of future applications will demand predictable latencies below 10 ms. This is the figure claimed by people working on remote Virtual Reality to avoid nausea and some telemedicine equipment designed for remote surgeries or rapid file transfer during emergencies. Autonomous vehicles need latencies closer to 1 ms. Four things become apparent from the material above. One, the server must be close, within a hundred miles or so. Two, the data rates have to be in the gigabit range in both directions. Three, there cannot be many switches and repeaters in the way. Four, server processing has to be quick. These system pieces do not exist now, but the applications are clearly in the wings; they will force data rates and latencies to improve considerably. From the perspective of the last-mile network, fiber optics will be a key component to make these applications a reality.
The transition from dial-up modems in 1978 at 2400 bps as the top speed to dial-up modems in 1998 at 56,000 bps as the top speed to DSL and cable modems running from 2 to 8 mbps as top speed in 2008 to cable modems and advanced forms of DSL running at 50 mbps top speed in 2018 will continue. At some point in the not too distant future cable and telephone networks, relying as they do upon copper lines, will be unable to keep up. Fiber optics all the way to the home or business will be the answer, with nominal data rates in both directions of 1 gigabit per second a kind of standard offering.