When worlds collide
2 - Circuit Switch telephony and Packet Switch networks
Telephones have been around a very long time. Alexander Graham Bell patented the telephone in 1876, so the telephone was over a century old before packet switch networking escaped from the laboratory. Hence the telephony industry is one of the most mature in the world of technology. In comparison, the computer industry is young and immature - if you were to anthropomorphise it you might see a troublesome teenager emerging from a difficult puberty. Asking the two to work closely together is bound to be interesting.
Digital Telephony Primer
Unless you work in the telecommunications industry your knowledge of how it works is likely to be quite limited, therefore it is worth describing the basics.
The analogue signal from your home or work phone is frequency limited to about 3kHz (very little content of speech is above this) and converted to digital using 8 bit samples, 8000 times a second, giving a bit stream of 64000 bits per second (64kbit/s). The telephone network consists of digital circuits capable of carrying 64kbit/s signals, and switches that can connect one circuit to another to form end-to-end paths. The basic building block of the telephony network is therefore the 64kbit/s data stream. An end-to-end circuit provides a dedicated path, it has a fixed latency, a fixed bandwidth, and a fixed quality of service. The resources allocated at every point in the network are fixed for the duration of the call. The number of circuits is also fixed so that once they are all in use all further calls have to be rejected; there is no option to adjust the balance between quality and numbers of calls.
Mobile networks are similar, although advances in processing power in the handset allow more efficient use of bandwidth in the access network so that the voice circuit between the handset and the core network only needs 16kbit/s.
Telephony networks also have a signalling infrastructure to control the calls carried over these 64kbit/s circuits. This is a message-based network designed specifically for call control and is often carried on the same physical media as the voice circuits. Initially this signalling was just restricted to basic call set-up and cleardown - passing the called and caller numbers and indicating when the call was answered and finished. As the number of services provided by the telephony networks has increased so has the number of type of messages carried by the signalling network so now it is far more than just call handling. The short message service (SMS) between mobile phones is an example of data carried by this signalling network that is not related to a telephony call.
Circuit Switch telephony networks
Telephony networks have developed slowly over more than a century, and until recently have been built, owned, and controlled by mostly state-owned monopolies. This long heritage has led to the characteristics of the modern telephony network that we are familiar with today:
· Dumb terminals. The modern telephone has changed surprisingly little since the 19th Century. It contains a microphone and speaker for the user to communicate with the other end of the call, and a method to signal to the network. The only significant change has been the replacement of the rotary dial with a keypad. You can take a hundred-year-old phone and use it to make a call to the latest model of mobile phone, or vice versa - there are very few other technologies that have remained compatible for such a long time. Even modern devices, such as the modem, facsimile, or DECT handset, are based on the same three components, transmit sound, receive sound, and a signalling interface.
· Intelligent networks. As services have been developed for the telephony world they have mostly been introduced into the core network. With the intelligence in the network, introduction and upgrade of a service does not require a change to the subscriber’s equipment, and the hardware, software, configuration, and security is under control of the network operator. This has been a great benefit to the network operators in the form of increased revenue, to the subscriber in the form of a rich menu of services, and to third parties in that they can sell services to everyone with a phone, no matter how basic.
· Metered charging. Calls are usually charged by time from the time of answer to the end of the call. This means that the cost of a call is transparent to the end user. Today’s networks support fixed price or unmetered charging, but these are a recent development and are only reluctantly implemented by network operators as they threaten existing revenue streams.
· Distance charging. Long distance calls have always cost more than short distance calls. This is based on the idea that the more pieces of equipment, or exchanges, the call passes through, the more expensive it is for the network operator, and that expense is passed to the end user. Additionally, interconnect charges are a valuable source of revenue so operators charge a premium for other operators to access their network, and therefore international and cross-network calls are relatively expensive.
· Fixed Bandwidth. The building block of the telecom network is the 64kbit/s channel. This has led to data connections using the circuit switch network reaching their limit at about 56kbit/s for analogue modems and 64kbit/s for end-to-end digital. Exceeding this limit requires more intelligence at the end terminals, for example, 128kbit/s ISDN is achieved by concatenating two 64kbit/s channels (charged by the Telco as two separate calls) and video calls by using 6x64kbit/s channels. Although there is some support in the network to route the concatenated channels via the same path, most of the functionality is implemented at the data terminal.
· Fixed Latency. The nature of the dedicated 64kbit/s channel is that every bit takes the same time to transverse the network, and therefore the latency for a connection remains constant. This is particularly important for voice communications as the human ear/brain is reasonably tolerant of delay, but not of jitter. The worst-case latency within a network in a medium sized country (e.g. BT in the UK) can be as little as 10-15ms.
· Partitioned Signalling. The circuit switch signalling network is partitioned into core network signalling and access network signalling. In the core network Signalling Scheme Number 7 (SS7 or C7) provides a trusted message transfer and relay mechanism. Access to the messaging in the core network is protected by having protocol conversion from an access network (typically ISDN signalling) that validates message content as well as providing supplementary services.
· Standards. The international telephony networks conform to nationally and internationally ratified standards from organisations such as ITU-T, ANSI, and ETSI. Standards are agreed at national or international level and then implemented by manufacturers and operators. The process is hierarchical and participation expensive, leading to domination by a small number of large organisations. There is also a historical global division resulting in a different set of standards in North America and (most of) the rest of the World.
Packet Switch networks
· Intelligent Terminals. Even the first networked computers were substantially more complex than a telephone. Even just providing the network interface required considerably more hardware and software than even the most advanced phones of the day. With intelligence in the terminal equipment, new services and features require modification to that equipment, resulting in costly and time-consuming upgrades. When a new service is rolled out, it is only available to those end users with the equipment that supports it, or those who are willing to upgrade their equipment.
· Dumb network. In packet switch networks the network is purely a transport mechanism. Most of the intelligence that seems to be in the network is actually provided by devices attached to the network, and each terminal device needs to know how to access network resources.
· Per packet or bandwidth charging. The charging model for access to packet switch networks traces its origins to the government (military) or academic use, which is often perceived as free. In practice, because most packet networks have a high infrastructure cost and low per-use costs, charging has been on a bandwidth basis, except where bandwidth is scarce when per-packet charging encourages efficient use. Where access to a packet switch network such as the Internet is provided by another network, such as dialup telecom access, the access charge is that of the access network.
· Distance independent charging. It is rare in the packet switch world to be charged by distance. Users of the Internet are often unaware of the location of the service they are accessing, so charging by distance could not be transparent and therefore would be unacceptable to most users.
· Flexible bandwidth. The usual method of sharing the bandwidth over packet switch networks is on a first-come first-served packet-by-packet basis. This means that a heavy user such as a batch file transfer will significantly affect a light traffic interactive user, and that the available bandwidth can vary during the lifetime of a connection.
· Variable Latency. Another consequence of the shared bandwidth is the variations in latency depending on the type of traffic sharing the bandwidth. This makes the transport of voice over contemporary packet switch networks a hit and miss affair.
· Transmission Delays. The packetising delay and anti-jitter buffering alone is often more than the end-to-end delay of circuit switch networks. There is a trade off between packet size and transmission delay as larger packets make more efficient use of the bandwidth at the cost of delayed transmission. Mixing bandwidth efficient traffic and latency-sensitive traffic on the same transport network requires more intelligence at intermediate nodes.
· End-to-end signalling. In most packet switch networks the intermediate nodes only deal with destination routing decisions, leaving all higher-level protocols to the endpoints. With the exception of HTTP, it is rare to validate the communication at any point within the network.
· Signalling Standards. The vast majority of packet switching protocols are defined by the IETF (Internet Engineering Task Force) via RFC (Request for comments) documents. This is an “implement first, then document what works” co-operative method of standardisation that allows anyone to partake without significant barriers to entry. Standards are mostly global although they tend to have a North American cultural and language bias.
The two worlds of circuit switch and packet switch are increasingly meeting and overlapping, and both are highlighting the other’s limitations when in the wrong domain:
· In the telecom arena the demands of packet over circuit is only possible by abandoning the current lucrative charging model and therefore disrupting the business model of traditional Telcos. The drive to extract higher transfer rates of data over the telephone access network has reached a limit at 56-64Kbits/s per circuit, and end users are demanding much more. The Telcos have always provided high bandwidth point-to-point circuits for business, but at a price. The charge for a 2Mbit/s E1 (1.5Mbit/s T1 in North America) is charged at thousands of pounds or dollars per month, depending on location and distance. Telcos have the choice of ignoring the demand and watching someone else take their market, or lowering the price and seeing the revenue stream dramatically reduce.
· The computer industry has until very recently always been forced to use the Telcos for anything other than on-campus connections. In many developed countries the stranglehold of the monopolies prevented companies from even linking two of their own buildings if they didn’t own the land between, no matter how narrow. This has made wide area networks expensive and time-consuming to install and run, leading to frustration and resentment of the Telco’s monopoly position. It is hardly surprising that the computer networking community, and more recently computer users, have taken every opportunity to circumvent or eliminate the Telco’s networks. At the extreme this has led to the technically crazy and inefficient use of voice over IP packet over digital voice circuit!
It is not surprising then, that the telecom industry is looking at moving its network infrastructure to packet switch, and the computer industry is increasingly providing services previously the exclusive domain of the Telcos.
For the telecom industry, packet switch allows them to satisfy the demands of IP based equipment connecting over their networks, carry traditional telecom traffic, and introduce new services to open up new sources of revenue. Unfortunately, the characteristics that make telecom networks reliable and predicable are mostly not present in packet networks, and retrofitting them is proving challenging.
Similarly, for the computer industry, retrofitting the reliable and predictable performance that has resulted from a carefully regulated telecom environment to their anarchical culture feels like a paradigm change too far.
Quality of Service (QoS)
The quality of service provided by the fixed line telecom networks in all developed countries is taken by granted by most users to such an extent that customers have become intolerant of dropped calls, lack of dial tone, and expect to get through first time, every time. Compare this with the computer industry where crashes are tolerated, perhaps almost expected, and modems dropping the line, missing web pages, and unavailability are taken for granted. The idea of a telephone exchange running for 25 years without a reboot is so foreign to the computer community that they have difficultly believing such things are possible. There is therefore a gulf between the expectations of telephone users and computer users, even when they are the same people. This gulf could be bridged by improving quality of service in computer networks up to that expected by telecom users, or reducing the expectations of the telecom users. Fortunately for the telecom industry, two technologies have become widespread in the last decade that have done much to lower the expectations of the telephone user: mobile phone networks and IVR (interactive voice response) systems fronting call centres. But can the quality of service in packet switch networks be improved significantly to allow them to match the now reduced expectations of the telephone user?
Voice over IP (VoIP)
VoIP is a much-misused term, often being confused with Internet Telephony. Although the technology has been around for many years, QoS issues are hampering its deployment. In order to provide a QoS comparable with the existing telecom networks, the underlying IP network needs to be carefully designed and managed, particularly with regard to capacity and shared traffic. In practice VoIP networks need to be grossly over-specified and dedicated to VoIP traffic, especially when using IPv4 (IPv6 solves some, but not all, of the QoS issues). It is simply not possible to just piggy-back voice traffic on an existing IP network and expect it to work reliably; either the network needs to be extensively upgraded, or a new VoIP network commissioned, both of which negate any perceived cost advantage of using an IP network. Despite the difficulties, VoIP is slowly being deployed, albeit mainly in two locations, PABXs, and IP islands in core telecom networks, both of which are controlled environments where QoS and other issues can be managed.
If a company needs to upgrade both its LAN and internal telephone infrastructure, there are significant cost-savings, both up-front and on going, by converging the two. Vendors offering solutions for the integrated corporate market will dominate a Google search for VoIP, and technical news feeds frequently relay press releases of companies making this change. Although this closed environment is a success story for VoIP, it is not the VoIP technology that is driving it, but the cost-savings from avoiding duplication, and often the biggest savings are in the physical wires rather than the protocols carried by them. It is highly likely that VoIP will be the dominant technology for providing voice services in the SME space by the end of the decade.
While in most of the VoIP world the publicity and hype precedes the implementation, the major telephone operators are quietly but steadily installing VoIP networks, if not as islands in their core network, at least in their test plant. My interpretation is that they believe VoIP is going to very important in the near future, but there are still significant technological, logistical and financial barriers still to break down, and that giving out too much information about the technology they are using would allow their competitors, especially the new entrants to the market, to piggy-back their research. Similarly, the major telecom equipment manufacturers, once you delve beyond the marketing hype, are vague about the direction they are taking, lest the big IP equipment manufacturers steal their market.
Most references to VoIP are actually referring to using the Internet for cheap telephone calls. On the surface this seems an easy way to avoid paying Telco charges, especially for long distance and international calls. In practice it is small niche application beset with problems. The variable, sometimes long, and unpredictable propagation delays of the Internet result in a low quality relegated to those who are willing to trade quality for low price. Unless the person you are calling is in one of the few areas with free Internet to telephone network gateways, you are limited to calling people who are close to their powered-up computer.
Despite these difficulties, a number of companies are providing Internet Telephony products. Noticeably the big players, such as BT with Broadband Voice in the UK, are by-passing existing desktop computers and supplying dedicated boxes plugged into the customer’s LAN and providing interfaces to the public telephone network, albeit not at zero cost.
VoIP is easy. Taking a digital voice stream, putting it into IP packets, and pulling it out at the other end is almost trivial, and the necessary supporting features such as codecs, jitter buffers, and echo cancellation are available both in hardware and as software algorithms. The element that makes VoIP usable but technically complicated is the signalling. Both the traditional telephony and the emerging IP telephony worlds have many copious standards, but whereas the traditional telephony world is in general agreement in which standards and options to apply where, there is no consensus in the IP world and each vendor is pushing its own favourite. Many of the IP telephony signalling protocols were developed before the current explosion in public internet use and the subsequent security and abuse problems, and as a consequence, do not co-exist well with the partitioning of the Internet such firewalls and NAT routers.
In its spring 1999 issue, the hacker magazine “2600” published an article entitled “SS7 explained” in which author Friedo describes in detail how SS7 works. He explains: “the hackability of SS7 does not at first appear possible, unless someone could figure out how to interface directly with the SS7 network”. Telecom service providers have been very protective about their internal system since the early 70s when John Draper discovered a toy whistle allowed users to circumvent billing systems for long-distance calls in the US and the resultant development of the so-called “blue boxes” sold to make it easy for end users to phreak the network. Even with the proliferation of mobile telephony networks and the licensing of many small operators, the security of the public telephone network is many orders of magnitude better than the Internet and the other IP networks connected to it. The public telephony network not only provides access to the emergency services but also provides many other critical links such as intruder detection alerts. The reliability and availability of the telephone network really is a life and death matter. It is simply not possible for IP based networks to replace the existing circuit switch telephony networks unless the security of IP is improved by orders of magnitude. This either requires an IP telephony network completely separate from the existing IP data network, which negates much of the advantage of an integrated system, or there needs to be a landmark change in how IP networks are deployed and secured.
There is already enough momentum in the direction that telephony and data networks are moving that by the end of the decade it will be impossible to tell where one network type ends and the other starts, and a time when there is no longer any concept of separate networks for telephony and computers is not far off. This brings great challenges all of us: The telephony world needs to shed the legacy of its monopoly position and gentle pace of technological change, particularly in the way it charges its customers and rations access to new technology. The computer industry needs to take a grown-up attitude to reliability, availability, and security; reboots, denial of service and viruses are simply not acceptable when dealing with universal public services and life and death situations. The most popular cliché at the moment is “wake-up call” - if you are in the telephony or data networks industry, this is yours.
telephony or data networks industry, this is yours.