This is a presentation on the evolution of the public network, including POTS, ATM, SONET, DWDM, RPR, Ethernet, and other technologies. Explains how needs and design principles have changed over time and compares the different technologies. A full research paper is available at http://www.ericgoldman.name
As new technologies are being developed, there is an increased awareness of the diverse needs of the public network. There is uniform traffic and bursty traffic, and traffic with various QoS requirements. The earlier technologies were designed with certain ideas about what the traffic would look like, whereas a more modern technology like RPR assume a greater diversity of traffic. New technologies will gain success only if they can be cost effective. For example both DWDM and RPR can reuse parts of the current infrastructure and are not overly complex or costly. In practice, these technologies do not operate in vacuums. One vendor may use different technologies were appropriate or may even combine technologies together to reap the individual benefits. While as IT people we always want the newest and the best, it is important to remember that the decision to use one technology or the other is more dependent upon costs and not always on perceived needs. As they say: If its not broken, we aren’t going to pay to fix it.
New protocol and technologies will continue to be developed as an increasing pace as our communication needs change. Currently, we do a lot of browsing on a web pages, but there is no guarantee that in 50 years all communications won’t be live or stored HD video, which will have a completely different set of requirements. However, our communications will always be limited by the infrastructure and thus we must continually work to get the most utilization out of what we have. The traffic currently on the public network is very diverse and different applications and traffic flows have different expectations and requirements such as QoS. An often overlooked problem is also that the public network involves a great number or parties, such as providers and end users, each with different goals and expectations, and no single decision is easy.
The public network is the aggregate of all the provider and backbone networks. This encompasses the Metro and Wide Area Networks. We use the public network every day to go online, watch TV, or talk to people on the phone. There are a lot of different types of traffic that must somehow find its way through this jungle of wires.Like most people, I usually do not worry about what happens in “the cloud”. However, as networked applications become more important a good understanding of the public network infrastructure can help a Sys Admin make better choice for the company. One day you may actually be faced with the question of purchasing leased lines or using a VPN over the Internet. Knowing what goes on inside the cloud would help you better address this question.
The good old POTS was relatively simple. Because the traffic was uniform and spikes could be predicted (for example Mother’s Day) it was much easier to manage. Also, there was less need for packet conversion since all data was traveling in a common format. However, the Internet changed everything. The amount of traffic over the network not only increased, but the traffic has become much more difficult to manage. Most packet based communications are short and bursty, for example loading a web page. You could also talk to different servers all over the world in a relatively short period of time. This greatly decreases the operators ability to predict traffic flows. In addition, more of the network is utilized at any given time, especially the more expensive long haul links. This was previously not a problem because long distance calling was expensive and thus not as frequent. Before the rise of the Internet, 20% of traffic was long haul and 80% local. Now, that ratio is reversed.
ATM was one of the first major widely distributed technologies deployed on the public network. It was designed in France for delivering cable television. ATM is connection-oriented which means that a path or circuit is reserved for some given data flow before transmission begins. Furthermore, there are built in provisions to ensure quality of service, including traffic policing (which ensures that if the traffic contracts are maintained or traffic is dropped) and traffic shaping (introduction of delay to ensure traffic conforms to specified profile). ATM uses a fixed-length packet which helps to reduce delay and jitter. ATM does not work as well as they designers hoped with Internet traffic. While it was hoped ATM would work well for data communications, it does not play very nicely with our mostly TCP/IP based traffic coming from subscriber networks. There is a mapping mismatch when Ethernet & TCP/IP is encapsulated over ATM which results in a cell tax for every 50 bytes of data. In addition, the process of mapping IP over ATM is difficult and expensive. ATM is still around because it works for the most part and providers do not want to loose their initial investment or pay for training until it becomes necessary.
SONET was designed as a generic carrier for both bursty and smooth traffic. Unlike ATM, SONET is not asynchronous and uses time division multiplexing. This approach allows for many providers to share a particular wire without worrying about the affects of another providers traffic load. In addition, when using asynchronous transmission bit stuffing and other costly overhead are avoided. SONET was designed for the public network and includes many provisions to provide network maintenance and administration with every frame as well as mechanisms for error correction. While many Optical multiplexing solutions will only operate in point-to-point, SONET allows for a wider range of topologies, with the most common being the ring. The Ring topology create redundancy and adds resiliency to the network. SONET is particularly useful on the public network because it is designed easily combine smaller payloads such as T1 or T3 which allows for these signals to be easily multiplexed together without breaking them down and reconstructing them later. SONET can handle very large payloads, with a base unit of about 51 megabits per second and easily exceeding gigabits per second. However, a SONET network tends to be very inflexible and does not always work well with the asynchronous nature of data communications.
We have already discussed the operation of DWDM in class, so I will focus on its application to the public network. First, it presents an increase in performance with a lower associated cost. The existing fiber used in SONET or ATM can be used with greater capacity in a DWDM system. In addition, DWDM equipment is less complex and less expensive that either SONET or ATM. One of the great benefits of DWDM is that it is an all optical process and can avoid conversion from light to electro-magnetic signals. The multiplexed signals can easily reach into the terabit per second range on a DWDM system. Since every carrier is independent and on a different wavelength, the processes of cleaning, boosting, and repeating a signal at an intermediary node becomes very complex. Another issues is that the data within any given DWDM carrier must be homogenous and destined for the same end point. This may often lead to underutilization.
RPR is one of the more intelligent technologies currently being considered for the public network. Its uses the current infrastructure of SONET which results in high resiliency, but is packet based in order to increase the performance of Ethernet and IP based transmissions. On top of this basic structure, all traffic is assigned to one of three traffic classes, which includes synchronous for real time transmission, guaranteed, and best effort. Traffic in an RRP networks travels on one of two active rings, whichever will have the lowest end-to-end cost for the transmission based on whether the ring is clockwise or counter clockwise. Compared with Ethernet, there is less processing of the traffic required as it travels through the network. One of the biggest strengths of RPR is spatial reuse. Unlike in a SONET system, when traffic reaches its egress from the RPR network that bandwidth can be reclaimed by another transmission increasing useful utilization. One of RPRs greatest strengths is its ease of integration with Ethernet networks, of which most subscriber networks are based.
We hear all the time that it would be great to have complete end-to-end Ethernet solutions. It would be much simpler to connect networks and there would be no conversion overheard for our data communications. However, Ethernet is just lacking in many areas. The connectionless orientation and lack of built in QoS make in inefficient for the public network, where standard voice is still the big money maker. In addition, Ethernet is most effective in a meshed architecture. The public network is comprised of rings and point-to-point links for which Ethernet would be inefficient. While Ethernet can now provide high speeds over fiber and there are mechanisms to implement QoS, these measures are not standardized or designed primarily for usage on the public network. There is no guarantee that these modifications exist end to end or are supported by all providers.