The Broadband Availability Gap: Traffic Modeling ChallengesThe Broadband Availability Gap: Traffic Modeling Challenges
Traffic engineering for converged networks is challenging--especially with the nature of bursty data traffic mixed with various real-time communications.
May 22, 2014
Traffic engineering for converged networks is challenging--especially with the nature of bursty data traffic mixed with various real-time communications.
The FCC paper "Broadband Availability Gap (OBI Technical Paper 1)," shows how traffic engineering for today's networks brings on new challenges not yet met by the scientific community on how to best model traffic that includes real-time communications across different media--wired, satellite, wireless, cellular etc.
The FCC states that:
"Modeling voice traffic makes use of the simple inputs of average duration of call, bits-per-second used by the voice encoding scheme and number of call originations per hour. This has enabled scientists and engineers over the years to develop reliable mathematical models that correlate well with real-world experience.
However, for Internet traffic, the number of variables, the magnitude of variation of these variables and the statistical nature of the variables have made it difficult for the scientific community to develop a well-accepted mathematical model that can predict network traffic based on end-user demand."
However, for Internet traffic, the number of variables, the magnitude of variation of these variables and the statistical nature of the variables have made it difficult for the scientific community to develop a well-accepted mathematical model that can predict network traffic based on end-user demand.
In past posts I've noted that traffic engineering for converged networks is challenging--especially with the nature of bursty data traffic mixed with various real-time communications: voice, video, conferencing, Skype, etc. New methods and models are needed that are based upon scientific approaches and not what many use to determine capacity--such as "averages" or "ratios"--because these primitive methods are not reliable and they do not work.
The FCC is using an average called Busy Hour Offered Load (BHOL):
"The data received/transmitted by a subscriber during an hour represent the network capacity demanded by the subscriber during that hour. This can be expressed as a data rate when the volume of data received/transmitted is divided by the time duration. BHOL per subscriber is the network capacity demand or offered load, averaged across all subscribers on the network, during the peak utilization hours of the network."
BHOL is akin to old voice modeling, in that we would examine one hour's traffic, and with really old models we would use the busiest hour of the busiest day of the year for the entire year, because right-sizing circuits was restricted by time and costs. In today's networks the striking difference is that data sessions do not last for minutes--applications can last for hours or just seconds and those hours or seconds may consist of bursty data.
Two important notes to consider when using any traffic model are that the PSTN is designed around a 99% call completion rate, and the cellular network is designed around a 95% call completion rate. Call completion also includes a busy signal because the dialed party is, in fact, busy. Then, another issue to add into the fold is latency, and each media used will involve more or less latency, e.g., satellite vs. fiber.
Then in their "Exhibit 4-BP" (shown below), the FCC notes the contrast between voice and data networks. These differences are what network engineers and administrators wrestle with daily in keeping converged networks running smoothly.
"User" and "customer" experience are on the agenda of many large enterprises, and I've seen the reports from the hospitality industry on their guest Wi-Fi ratings of their properties. Wi-Fi presents another challenge--and even unique challenges within each industry that adopts Wi-Fi.
BHOL, for the most part, represents an average. New and improved methods of traffic engineering for converged networks are needed, and this means that until they are developed and proven, capacity planning may need revisiting. Perhaps these criteria may leave some doubt, so add in the fact that Comcast and other providers want to limit bandwidth usage and charge in Tiers or for overages of data limits.
Will public network performance diminish because of BHOL averaging and hooks placed by the providers? Will "blocking" become the new norm for networks, with user/customer experience thrown out the door? Maybe the term "network" is becoming as ambiguous as the term "cloud".
The ramifications are challenges to both large enterprises and SMBs. While the large enterprise remains somewhat protected within private networks, the SMBs remain largely on the public network best -effort route across the Internet. Both grade of service (GoS) and quality of service (QoS) may be impacted in the convergence effort as the PSTN fades.
Follow Matt Brunk on Twitter and Google+!
@telecomworx
Matt Brunk on Google+
