9.4 Case Study: Asynchronous Transfer Mode
Historically, there have been different networks that carry different types of information:
• Telex (old style teletypewriters, used for news feeds, stock quotes, etc.)
• “Plain old telephone service” (POTS) via the public switched telephone network (PSTN);
• Data, via the packet switched data network (PSDN);
• Television via: (1) ground based antennae; (2) community antennae TV (CATV known as cable TV); and (3) satellites;
• Local data via local area networks (LANs).
Each network is separately planned, dimensioned, developed, and operated. This specialization results in independent worldwide networks. Bandwidth is not shared among the networks, yet when video traffic peaks at prime time, telephone traffic lulls. Although there may always be video programs available on a particular channel at all hours of the day, less-expensive local programming may be used in the off-peak hours, reducing the need for video traffic via satellites or via long distance wireline networks.
Each type of service needs to deal with peak traffic and bursty traffic on its own, although it would be more economical to share networks as long as the peaks and bursts do not coincide. An attempt at sharing is Integrated Services Digital Network (ISDN), which comes in two forms. Narrowband-ISDN (NISDN) is designed for 64 Kb/s telephone switching. NISDN allows voice and data traffic to be carried over the same network. Broadband-ISDN (BISDN) supports video traffic as well as voice and data traffic.
The ISDN Basic Rate Interface (BRI) is used for NISDN. The BRI offers two Bearer (B) channels at 64 Kb/s per channel, for user data, and one Data (D) channel at 16 Kb/s which is used for control and signaling. The total bit rate is 192 Kb/s when higher level framing overhead is taken into account, but the maximum rate available to the user is 128 Kb/s when the two B channels are “bonded” into a single channel.
The ISDN Primary Rate Interface (PRI) offers 23 B channels and one 64 Kb/s D channel, and is commonly known as a “T1 line.” ISDN lines can be leased from telecommunication companies and can be configured to set up private networks. As a general rule of thumb, the monthly rate for leasing 10 64 Kb/s lines equals the rate for leasing a T1 line that supports 23 64 Kb/s lines. A problem with the overall economics with ISDN is that the sup- ported bit rates are service dependent (NISDN is an example of a service) as opposed to traffic dependent, which is needed by modern networks. Services must fall into even multiples of B and D channels in the offered ratios, or else they will not make efficient use of the available bandwidth. We may not know what bit rates future services will need, and so we need a service independent bit transport for a network to be extensible. The goal is to have a single network that can serve everyone, which is where asynchronous transfer mode (ATM) comes in.
9.4.1 SYNCHRONOUS VS. ASYNCHRONOUS TRANSFER MODE
With synchronous transfer mode (STM), also known as time division multiplexing (TDM), a data stream is made up of time slots that are assigned to channels in a round robin fashion. Figure 9-20a illustrates TDM for four channels. A station can only send during a preassigned time slot. Other unused time slots may go by while a station waits for its preassigned time slot. (In the public switched telephone network, a time slot is 125 µs, which allows for 8000 samples per second at 8 bits per sample, resulting in 64 Kb/s for a voice grade line.)
Figure 9-20b shows asynchronous transfer mode, which may still use 125 µs framing, but now, any station can use any slot, and the network is used more efficiently. On the down side, operating an ATM network is much more complex than operating a simpler TDM network.
9.4.2 WHAT IS ATM?
ATM is a combination of hardware and a set of protocols that delivers a guaranteed bandwidth with a bounded low latency. Two enabling progress areas that make ATM possible are:
1) Technology – The speed and density of VLSI allows network functionality to be pushed into the end-systems rather than throughout the intermediary components. Error detection is only performed at the endpoints, for example, instead of at every intermediate router as is common for the Internet. The development of high bandwidth, low bit-error rate optical fiber is a supporting technology that makes this approach practical.
2) Systems – Fast packet switches that introduce a low latency into end-to-end communication enable the delivery of real-time services.
Typical speeds for ATM are in the range of 1 Gb/s – 2.5 Gb/s, with an average delay of only 450 µs for each switch that handles an ATM packet.
9.4.3 ATM NETWORK ARCHITECTURE
Before an ATM transfer takes place, a connection must be established by contacting a signaling control point (SCP) which is a network device that has the authority to configure ATM switches for the transfer. All packets are constrained to follow the path set up by the SCP, and all packets arrive in order.
The job of the switch is very simple: it simply looks at the destination address in the header, and then it sends the packet on the path indicated in the header. A new destination address is placed in the header, as described further below.
All ATM packets have the same fixed size of 53 bytes, as shown in Figure 9-21.
The packet created by an end system for injection into the network, which is known as the user-to-network interface (UNI) format is shown in Figure 9-21a, and the packet format used from that point onward, which is known as the network-to-network interface (NNI) format is shown in Figure 9-21b.
The generic flow control (GFC) field is used at the network boundary to police how packets are allowed to enter the network. Once a packet is in the network, the GFC field of the UNI format is no longer needed, and is then combined with the 8-bit virtual path identifier (VPI) field to form a 12-bit VPI field in the NNI format. The VPI field can be thought of as identifying one particular home for a cable that carries cable TV channels, whereas the virtual channel identifier (VCI) field identifies the specific channel. The payload type identifier (PTI) identifies whether the data field carries user data or network data, and other information. The cell loss priority (CLP) bit determines whether this cell can be dropped during times of congestion. The header error control (HEC) is a CRC over the header. The 48-byte payload field follows.
Figure 9-22 shows a simple ATM network that consists of three switches. Each
switch has a symmetric routing table that can be driven in either direction. For example, at ATM Switch #1, an incoming packet (on the left) that has a VPI field of 7 is sent to the right after changing the VPI field to 5. Likewise, a packet that comes into ATM Switch #1 from the right with a VPI field of 5 is sent to the left after changing the VPI field to 7. This is referred to as “virtual path switching” because only the VPI field is used in routing the packet. The VCI field is unchanged. There can also be “virtual channel switching” in which routing is done based on the VCI field, and the VPI field is left unchanged.
9.4.4 OUTLOOK ON ATM
Although ATM holds a great deal of promise for an extensible network that serves many user needs, economies of scale favor legacy networks such as Ether- net for end-user systems. ATM appears mostly in backbone networks, with multiplexors and concentrators used for connecting legacy networks at the boundary of the backbone to the backbone itself. ATM continues its penetration into back- bone networks and also to the desktop, which is currently the exception rather than the rule. Whether ATM makes it to the desktop pervasively, time will tell.