11/20/2002 -- If you have a background in packet-switched data communications, many of the terms and concepts used in the traditional voice telecommunications world are foreign. In an earlier article, I gave an overview of wide area networks (WANs) that touched on SONET as the principal backbone technology for common-carrier WANs. A few years ago SONET may have been considered to have a limited future as all-Ethernet and other packet-based overlay networks took over the world. The economics of the game have changed, however, and now it looks as if SONET and its successors will be with us for the foreseeable future.

This article is designed to give a more in-depth look at SONET architecture and its unique terms in a way that those with a previous understanding of datacomm networks can comprehend. If you've spent some time trying to do this research yourself, you know that most articles on SONET read like incomprehensible gibberish, so I'll try to provide a telecom/datacom Rosetta Stone to help bridge the gap.

The Background
Traditional digital telecommunications services such as T1/DS1s were designed to aggregate analog telephone lines for more efficient transport between central offices. Twenty four digitized voice lines (DS0s) were carried over a DS1 using time-division multiplexing (TDM).

To review, in a TDM architecture, multiple channels -- in this case 24 -- share the circuit basically in rotation, with each DS0 having its own assigned time slot to use or not as the case may be. As each channel is always found in the same place, no address is needed to extract (demultiplex) that channel at the destination. 28 DS1s are TDM aggregated into a DS3 in the same manner. Contrast this with a packet-based network such as Ethernet where there are no assigned time slots; instead, the packets are multiplexed onto the media via an arbitration mechanism (CSMA/CD) and demultiplexed via MAC address.

The older DS1/DS3 system is known as the Plesiochronous Digital Hierarchy (PDH), as the timing of signals across the network is plesiochronous, which means almost but not precisely synchronous (now you have a new word to use in cocktail conversation). Data communications networks such as Ethernet are asynchronous, as there is not a centralized timing source and each node has its own clock.

As more and more channels are multiplexed together into higher layers of the PDH hierarchy, a number of problems arise. Since the timing on various DS1s going into a DS3 may differ slightly, padding of some channels (bit-stuffing) is required to align all within the DS3 frame. Once this is done, the individual DS1s are no longer visible unless the DS3 is completely demultiplexed. In other words, you cannot just pick off an individual channel but must tear the whole DS3 frame down, take the DS1 you want out, then rebuild the DS3; the equipment required to do this is expensive. Another problem comes where different networks with relatively wide differences in timing meet, such as Europe and the U.S.; expensive equipment that also adds latency is required for the interface.

To alleviate these problems, in the mid-'80s the Synchronous Optical NETwork (SONET) was designed. The International Telecommunications Union later generalized SONET into the Synchronous Digital Hierarchy (SDH) in order to accommodate the PDH rates in use outside North America.

SONET Layers
Each SONET network node ultimately derives its timing from an exceedingly precise and stable cesium atomic clock somewhere on the network, leading to the "synchronous" part of the name. The SONET model is totally different from the familiar OSI datacomm 7-layer model. The SONET model defines four layers:

• Photonic, which corresponds to the OSI's physical layer, defines the optical equipment's attributes (OC-n.) The tolerances in areas such as signal timing, jitter, phase shift, etc. required to maintain a synchronous network over a wide area are exacting, much more so than in asynchronous networks such as Ethernet, and we won't get into them here; those interested can order up the inches-thick Bellcore specifications for their reading enjoyment.
• Section, the frame format and certain low-level signal definitions, roughly corresponding to the OSI link layer.
• Line, the way in which lower-level frames are synchronized and combined into higher levels; you can sort of look at this as parts of the network and transport layers. The line layer also defines data channels carrying operations, administration, maintenance and provisioning (OAM&P) information, which would be an application layer (like SNMP) in an OSI modeled network.
• Path, the end-to-end transport of a circuit, which also has application information (performance monitoring, status, tracing) for management.

The section, line and path layers correspond to types of equipment (known as network elements) as shown in Figure 1, above. A basic element is the SONET terminal, or terminal multiplexer, which concentrates DS1s or other non-SONET signals into a SONET link.

A very simple SONET network could consist of two terminals with length of fiber between them. If the distance is too long for one fiber link, regenerators are used to amplify and reconstruct the physical signal. An add/drop multiplexer provides two fiber connections with the ability to access the internal structure of the SONET frame to remove or insert individual channels as required for that node while passing the rest of the traffic on through. Cross-connects are used to switch, combine, redirect, and otherwise groom traffic, with varying degrees of granularity. Some operate only on the STS-n level (see below), while others can switch or multiplex down to the VT (see below) or DSn level. All of these elements are section terminating equipment; all except regenerators are line terminating equipment. Network elements where non-SONET signals are attached to the SONET network are path terminating equipment. All elements are intelligent, accessing in-band management information dedicated to each layer within the SONET frame.

Within metropolitan areas, SONET networks are typically configured physically as rings (see Figure 2, above). A ring topology provides a single level of redundancy, allowing restoration of service if one fiber link is broken. The SONET mechanism for restoration takes less than 50 milliseconds to recover from a break, but is considered somewhat inefficient as half the total ring bandwidth is reserved. Note that even though the physical topology may be a ring, the individual channels (which are manually provisioned) are point-to-point -- SONET has no equivalent of Ethernet/IP broadcast or multicast service.

The STS-1 Frame
The basic building block of the SONET protocol is the Synchronous Transport Signal level 1 (STS-1). Unlike Ethernet or IP where the frame structure is usually illustrated linearly, the large frame sizes involved in SONET are depicted as two dimensional matrices. The STS-1 frame is illustrated as an array of bytes 90 columns wide by nine rows high. Read left to right, then top to bottom, this works out to 810 bytes or 6480 bits per frame, transmitted at a rate of 8,000 frames per second resulting in the basic SONET signal rate of 51.840 Mbit/sec. All higher level signals are multiples of this rate. Note that although the STS-n hierarchy defines the signal rates, these are carried on optical fiber using the equivalent OC-n hierarchy.

The STS-1 frame consists of overhead and payload. The first three columns comprise the transport overhead; the remaining 87 columns are called the Synchronous Payload Envelope (SPE) (see Figure 3, below). The transport overhead section has the framing, performance monitoring, pointers, alarms and other OAM&P information used by the section and line layers. Within the SPE of 9 rows by 87 columns is the path overhead, found in column 1, and what is known as "fixed stuff" in columns 30 and 59; this has end-to-end monitoring and more performance information. The remaining 84 columns (756 bytes) are the revenue-generating part of the whole frame, known as the STS-1 Payload Capacity. A little quick math shows that the total overhead of six columns (three transport and three SPE overhead) within an STS-1 frame is 9.5 percent; looked at another way, the 51.840 Mbit/sec signal can carry 48.38 Mbit/sec of payload.

While it is easiest to envision the Synchronous Payload Envelope as a 9×87 rectangle beginning in the top row of the STS-1 frame just after the three bytes of transport overhead as shown in Figure 2, in fact it can start anywhere in the frame and extend into the next with the SPE start indicated by a pointer (see Figure 4, below).

For higher levels in the STS-n hierarchy, STS-1 building blocks are byte-interleaved to produce an STS-n frame consisting of n times 90 columns by 9 rows, still transmitted at 8,000 frames per second. For instance, an STS-3 has 270 columns × 9 rows resulting in a signal rate of 155.52 Mbit/sec. The overhead columns of each STS-1 are aligned with the STS-n frame, but the SPE payloads need not be as pointers indicate the start of each. The STS-1s may be independent of one another, or concatenated to provide greater channel capacity than the basic building block provides. Concatenated frames are indicated by a "c" suffix; an STS-3c has three concatenated STS-1s for a payload capacity of 145.15 Mbit/sec. Higher levels may also be concatenated in multiples of STS-3c's.

It is important to always remember the big difference between time division multiplexed networks such as SONET and packet-switched networks such as Ethernet. In a packet network, a node only packages up a frame and sends it out when there is some sort of payload; the wire may be "dead" between packets. In SONET, the frames are sent back-to-back, continuously, regardless of whether the payload portion is occupied or not. Visualize the STS-1 frame as a long (and really fast!) freight train. It has three engines (transport overhead), then 87 freight cars (payload envelope), then another three engines, 87 more freight cars, etc. for a total length of 810 combined engines and freight cars. Within each group of 87 freight cars the 1st, 30th, and 59th (path overhead) are used by the railroad and not available to carry freight. This train is moving really fast, with all 810 cars passing by in 125 µsec. Hitched up to the last of the 810 cars in the train is another one that is identically configured, and so on to infinity.

We can even extend this analogy to higher levels in the SONET hierarchy. Three of these infinitely long and very fast STS-1 trains converge on a switchyard. Both engines and freight cars on the first are all painted red, the second all white, and the third all blue. When the trains reach the switchyard, the first engine from each of the trains is unhitched, then the three instantly interleaved (multiplexed) into a single train leaving the switchyard; then follow the second and third engines from each and the 87 freight cars. So what you have leaving is a train three times as long, starting with 9 engines sequentially red/white/blue/red/white/blue/red/white/blue followed by 783 freight cars alternating colors the same way. Then come the next nine engines, and so on. This sequence repeats nine times to complete the STS-3 express freight train, 2,430 total engines and cars -- and note that it still takes the same 125 µsec for the longer train to pass, so each car is moving three times as fast. This shows the necessity for exact synchronism of the three feeder STS-1 trains, so that the interleaving can take place without a train wreck! And note also that each of the original trains' cars may be located within the longer train, as they are always at a fixed position from the front; down the line another of our hypothetical switchyards can readily disassemble (demultiplex) the STS-3 into three STS-1s going to different destinations. The rest of the SONET hierarchy can be built up the same way, with four STS-3s being combined into an STS-12, four STS-12s into an STS-48, and so on. In every case, the time for one train (frame) to pass by remains 125 µsec while the train length multiplies so the speed of each car increases proportionately.

Paying the Freight
O.K., so an STS-1 payload can handily accommodate one 45 Mbit/sec DS3. As a DS3 carries 672 voice calls, this is a pretty coarse level of granularity. However, the STS-1 payload can also be divided up into a number of Virtual Tributaries (VTs) ( known as Virtual Containers, VCs in SDH terminology) designed to provide synchronous transport for lower-speed PDH channels. Virtual tributaries come in three sizes corresponding to three PDH signals:

Each STS-1 SPE payload can carry a mix of virtual tributary types. The SPE is divided into seven VT groups of 12 columns each; each group may contain four VT-1.5s, three VT-2s or one VT-6 (but not mixed within the group). Hence if all DS1s are involved, an STS-1 can carry 4×7=28 DS1s. Because of the way VTs utilize defined column locations within the SPE, individual VTs can be identified and picked out without breaking down the entire STS-1. Contrast that with the plesiochronous DS3, where the entire frame must be broken down and reassembled if a DS1 is to be extracted.

Within the virtual tributary, there is VT path overhead and VT payload. In order to map the plesiochronous signals into the synchronous virtual tributary, pointers are used so that the tributary signals may float within the VT. Also null bits may be added ("stuffed") or removed to accommodate differing frequencies or jitter. Because of this flexibility within the rigid boundaries of the STS-1 frame, it seems to me that the SDH "Virtual Container" name is a more apt description. To take our freight train analogy to an extreme, the VT (or VC) is a set of tanker cars located at the same position in each STS-1 train; as the gauges on the pumps that fill each tanker car are not quite accurate, the pumps are set to fill each car not quite to capacity to allow for this variance.

So That's SONET...
To summarize the key attributes of SONET:

• The whole network is synchronous, which requires physical layer timing much more exacting than asynchronous data networks.
• It is a time division multiplexed network, where each individual channel occupies a fixed position within a continuous stream of frames.
• A hierarchy of rates starts with a 51.84 Mbit/second STS-1 frame. Higher levels are formed by interleaving or concatenating STS-1s into STS-n frames. Each STS-n frame takes 125 µsec, so the bit rate increases up the hierarchy by n times.
• Within the frames there are payload and overhead sections, where the overhead includes elements of what in the datacomm world would be link, network, transport, and application layers.
• Outside North America, SDH rather than SONET is deployed; SONET can be considered a subset of SDH, and the two can interoperate.

While it may seem we covered a lot of details in this article, in reality it is just the tip of the iceberg of the SONET specification. Say what you may about the business model of the old AT&T system, they certainly could design things right, to an exacting level of specificity. Now, one thing that was not considered at the time was how rapidly data traffic would grow, and SONET is not particularly good at carrying packet-based traffic. For example, an entire STS-1 must be allocated in order to handle a 10 Mbit/sec Ethernet channel. But extensions to SONET are now being developed and implemented which greatly improve its efficiency at carrying bursty data, as it looks as if SONET will be around for a long time.

Questions? Comments? Post your thoughts below!

More articles by Randy Bird:

• Video Extension Technologies 101: A Primer
  
• Emerging Optical Technologies: Resilient Packet Ring and Passive Optical Networks
  
• Cranking It Up Again: 10 Gigabit Ethernet Arrives
  
• Understanding WANs: A Technical Primer