Outline

• Finish Up Links
  • Leased Lines
  • Last-Mile Links
  • Wireless Links
  • NRZ, NRZI, Manchester, 4B/5B Encoding
  • Framing
• Signal Encodings

Introduction

• On Monday we began talking about Direct Link Networks.
• We looked at the hardware involved in network nodes.
• We discussed how the CPU communicates with the network adapter using the CSR and can send and receive using either DMA or PIO.
• We made the point that CPU speeds are increasing faster than memory access speeds, as well as bus speeds, and this can limit our ability to send data as fast as we might want.
• Also, it shows we don't want to perform too many reads or write to memory in transferring data.
• Another issue we discussed is that our memory is finite and so this limits our ability to buffer on a given machine.
• We also started talking about links, frequency, wavelength, and the EM spectrum.
• We now look more into the details of implementing links...

Leased-Lines

• If the two nodes you want to connect are on the opposite sides of the country, or even across town, it is not practical to install the link yourself.
• You could lease such a dedicated link from the phone company called a "leased line".
• Below is a table of the various "leased line" services the phone company might provide:

  Service	Bandwidth
  DS-1	1.544 Mbps
  DS-3	44.736 Mbps
  STS-1	51.840 Mbps
  STS-3	155.250 Mbps
  STS-12	622.080 Mbps
  STS-24	1.244160 Gbps
  STS-48	2.488320 Gbps
  STS-192	9.621504 Gbps
  STS-768	38.486016 Gbps

More on Leased-Lines

• DS1 and DS3 (sometimes called T1 and T3, DS := digital signal) are older technologies originally defined for copper-based transmission media.
• DS1 is equal to the aggregation of 24 digital voice circuits of 64kbps each and DS3 is equal to an aggregration of 28 DS1 links.
• All the STS-N links are optical fiber (STS stands for Synchronous Transport Signal).
• STS-1 is the base link speed and STS-N has N times the bandwidth of STS-1.
• STS-N is sometimes called OC-N (optical carrier).
• Typically, the phone company does not implement a leased link using a single cable, but instead uses its own internal network.

Last-mile Links

• Instead of a dedicated leased line, it is often cheaper to connect two hosts by connecting each of them to a network using a last-mile link.
• Some last-mile services are (some ancient, some more recent): modem (POTS - plain old telephone service -- ancient), Integrated Services Digital Network (ISDN, ancient) - two 64Kbps channels stuck together one for voice the other for data (can ditch voice if want), xDSL (some flavor of digital subscriber line ADSL or VDSL) can go over POTS but where the voice filter on local loop has been removed at end office, G.Fast is a DSL technology often used in apartments, cable, or passive optical network.
• Below is a table of common last-mile links and their bandwidths:

  Service	Bandwidths
  POTS	28.8-56Kbps
  ISDN	64-128Kbps
  xDSL	128Kbps-100Mbps
  G.Fast	up to-1Gbps
  Cable (depends on number of channels)	around 1 Gbps
  PON	up to 10Gbps

Wireless Links

• Wireless links transmit electromechanical signals - radio, microwave, infrared, or even optical light (optical telegraphy was used by Napoleon).
• The space through which these signals are transmitted is shared.
• Since signal strength falls off with the square of the distance, a given source attenuates with distance and this can be used to limit who hears who in space.
• The other way we can control who uses this common resource is by coming up with protocols about which frequency to send data on.
• In the U.S., the FCC (Federal Communications Commission) is responsible for reserving certain frequency bands for certain uses: AM radio, FM radio, TV, etc.
• It also licenses specific frequencies within these bands to organizations for a given area. For example, KQED.
• There are license-exempt frequencies that have different restrictions on their use: low power, etc.
• There are several techniques for sharing a limited spectrum of frequencies among many users so as to reduce interference (spread spectrum techniques):
  • frequency hopping -- sender and receiver share a pseudo-random number generator (prng) seed, and use this seed to compute a sequence of frequencies to briefly communicate on.
  • direct-sequence -- each bit of data is represented by multiple bits in the transmitted signal, so if some bits are damaged by interference it is often still possible to figure out if the code was for a 1 or a 0. For each bit the sender wants to transmit, the sender actually XORs that bit with n random bits generated by a (prng) known to the sender and receiver. The transmitted value is known as an n-bit chipping code.
• Spread spectrum techniques are used with 802.11 (Wi-fi), 802.15.1 (Bluetooth), 802.16(Wi-Max).

What is the connection between frequency and channel capacity?

• The Shannon Hartley Theorem gives the following upper bound on the number of bits per second we can communicate without error across a bandwidth `B` (in Hz) channel with signal power `S` and Gaussian noise power `N`:
  `C = B \log_2 (1 +S/N)`
• For example, on a 1990s era telephone line typically supports frequencies between 300Hz and 3000Hz, so has a bandwith of 3000Hz. The ratio of S/N was around 1000, so `C = 3000 times log_2(1001) approx 30`kbps.

Encoding

• We want to be able to write 0 and 1's using our ability to send the signals low and high between machines.

NRZ Encoding

In the nonreturn to zero (NRZ) encoding we use high for a certain amount of time to represent a 1 and low for a certain amount of time to represent 0.

• The receiver usually keeps an average of the signal seen so far, so if several consecutive 0's or 1's are sent it skews the average, making it hard to distinguish between 0's and 1's. (baseline wander)
• The sender and receiver need to synchronize their clocks in order to know how long a single 0 or 1 is. To do this the receiver typically relies on seeing 0 to 1 or 1 to 0 transitions from the sender so it can align its clock. However, if too many 0's or 1's in a row are sent this is not possible.

NRZI Encoding

In the nonreturn to zero inverted (NRZI) encoding, to send a 1 the sender transitions from the current signal value to the opposite; to send a 0 the sender stays on the current signal.

• This solves the problem of consecutive 1's, but does not help with consecutive 0's.

Manchester Encoding

In the Manchester encoding, one sends the XOR of the clock and the NRZ-encode data.

• This results in a 0 being encoded as a low to high transition and a 1 being encoded as a high to low transition.
• The drawback is the rate at which the signal changes (the baud rate) to encode a single bit has doubled.
• The bit rate/baud rate is only 50% which is somewhat inefficient.

In-Class Exercise

• Draw a picture of Manchester encoding for the following bit patterns 000000, 111111, 010101, 101010.
• Please post your solution to the Feb 8 In-Class Exercise Thread.

4B/5B encoding

• In 4B/5B encoding, every 4 bits of data are encoded in a 5-bit code sent to the receiver (who must then decode them).
• These codes are sent using NRZI.
• The 5-bit codes are selected such that each one has no more than one leading 0 and no more than two trailing 0s.
• So when sent back-to-back, no pair of 5 bit codes results in more than three consecutive 0s.
• The bit rate/baud rate efficiency of the code is 80% -- so is better than Manchester encoding.
• The Wikipedia page linked to above has a table of the 5-bit codes. For instance, 0000 maps to 11110.
• There are 32, five bit strings, 16 are used for data, 7 are not valid, and six are used for control values. One also has 11111 -- line is idle, 00000 - line is dead, and 00100 -- halt.
• 4B/5B is used by FDDI optical ring networks which we'll describe later, it is also used by Fast Ethernet.
• There is also a 8B/10B scheme which uses the same idea applied to 8 versus 10 bits used by PCI Express and Firewire, a 64B/66B scheme used by 10GB Ethernet, and 128B/130B scheme use by PCIe, SATA 3.2, DisplayPort 2.0, USB 3.2 and 4.

Framing

• Let's assume we can now send a sequence of bits over a link.
• Since we will be focusing on packet-switched networks, we are interested now in splitting this bit-stream into blocks of data called frames.
• We will first look at this problem for point-to-point links and then for multiple access networks.

Byte-oriented Protocols

• A protocol which views each frame as a collection of bytes rather than bits is called a byte-oriented protocol.
• Such protocols were originally developed to connect terminals to a mainframe.
• For example, Binary Synchronous Communication (BISYNC) was developed at IBM and Digital Data Communication Message Protocol (DDCMP) was developed at DEC for this purpose.
• Point-to-point (PPP) is a more recent and widely used example of this approach.
• We'll now discuss two types of byte-oriented framing: sentinel-based framing and byte-count framing.

Sentinel-based Framing

• BISYNC uses special characters called sentinel characters to indicate where frames start and end.
• The SYN character is used to indicate the start of a frame, there is a SOH character to indicate a start of a frame header, an STX character to indicate start of text, and an ETX character used to indicate the end of text.
• A BISYNC frame ends with a CRC (cyclic redundancy check) character used to detect errors in the frame.
• BISYNC uses character stuffing (preceding a control character with a special data-link escape (DLE) character) if a SYN, ETX, etc need to appear in the text or header.

PPP Frames

• PPP is commonly used to carry internet packets over point-to-point links (could be modem; could be long-haul links).
• It was defined in RFC 1661
• Flag is a sentinel with bit pattern 01111110; Address and Control usually contain default values 0xFF and 0x03; Protocol is used to identify the high level protocol such as IP or IPX(Novell).
• The Payload can be up 1500 bytes and character stuffing might be used in it.

Link Control Protocol

• The Protocol field is negotiated using the Link Control Protocol (LCP).
• LCP actually makes use of PPP frames to send control messages to do this negotiation. These messages have as their protocol a special LCP identifier.
• LCP is also used to establish a link between peers when both sides detect that communication over the link is possible.

Byte-count Framing

• In the byte-counting approach, one sends the number of characters in the frame at the start of the frame.
• DDCMP uses this approach. Its frames look like:

  8 bits	8 bits	8 bits	14 bits	42 bits		16 bits
  SYN	SYN	Class	Count	Header	Body	CRC

• So called framing errors occur with both byte-count framing and sentinel framing if the COUNT field is incorrect or the ETX field is corrupted. These can be caught using the CRC field.

Bit-oriented Protocols (HDLC)

• The high-level data link control protocol (HDLC) is an example of a bit-oriented protocol -- the payload of a frame is viewed as a sequence of bits which might not necessarily corresponds to bytes in some character encoding.
• HDLC is an ANSI and ISO standard derived from the SLDC -- Synchronous Data Link Control protocol used in IBM mainframes.
• A HDLC frame looks like:

  8 bits	16 bits		16 bits	8 bits
  01111110	Header	Body	CRC	01111110

Bit-stuffing

• If the beginning sequence 01111110 is needed within the body of the message a technique known as bit-stuffing is used:
  • Whenever 11111 is sent within a frame, the sender immediately sends a 0, before sending the next bit.
  • The receiver when receiving these extra 0's immediately removes them from the decoded bit stream.
• Notice for both bit-stuffing and character stuffing, the size of the frame depends on what is being sent.

SONET and Clock-based Framing

• Sychronous Optical Networks (SONET) use a third approach to framing known as clock-based framing.
• SONET was developed at Bell Communication Corporation, ANSI, and it has been adopted by the ITU-T.
• SONET solves both the framing and encoding problem and it also supports multiplexing several low-speed links onto higher speed links.
• In SONET a frame is sent every 125μS whether or not one has data. Frames are sent based only on the clock so are synchronous.
• One kind of traffic we'll later consider over SONET is called ATM. ATM cells do not arrive based on a clock, so the A in ATM stands for asynchronous.

SONET STS-1 Frames

• An STS-1 frame consists of 9 rows each of 90 bytes.
• The first 3 bytes of each row are overhead, but the remaining "payload" bytes can be used for data.
• The first three rows of overhead data are called the section overhead. Section overhead is checked at the start of each section (frame).
• As an example the first two bytes of each frame (first two overhead bytes of first row, i.e., section overhead) contain a special bit pattern.
• To sync itself with the sender, the receiver looks for this bit pattern repeating every 810 bytes.
• The last six rows of overhead bytes are called line overhead. They are checked each line rather than each frame.