COMP 3400 Networks Lecture 8: Data Link Layer
major resource: Computer Networking (4th Edition), Kurose and Ross, Addison Wesley, 2008

[ previous | schedule | next ]

Overview

• alternately called data link layer or link layer
• between network and physical layer
• partitioned into two sublayers:
  • logic link control (LLC) technology-independent "upper" sublayer
  • media access control (MAC) technology-dependent "lower" sublayer
• unit of transmission is called a frame
• focus on communication between 2 "directly" connected nodes
• Link layer is normally associated with LANs (Local Area Networks)
• "directly" is in quotes above because conceptually the LAN is broadcast but physically it may be segmented

Major services/issues

• Framing - packaging the payload
• Channel access - since channel is likely broadcast
• reliable delivery - where have we seen this before?
• flow control - we've seen this issue before too!
• error detection and correction - this too!
• Most link protocols implemented in hardware
• Host device is Network Interface Card (NIC), aka network adaptor
• Network extending devices include: repeaters, hubs, switches, bridges
• Focus on IEEE 802.x standards for LANs

Error Detection and Correction

• signal errors occur as result of interference, attenuation, etc
• all error detection schemes require redundancy (extra bits)
• Major error detection techniques include:
  • checksum - we've already seen this in UDP, TCP, and IP
  • parity - extra bit(s) to make count of 1 bits odd or even (can be fooled by even number of bit errors)
  • CRC - cyclic redundancy check
    • CRC bits appended such that combination of data and CRC bits form value divisible by "well-known" divisor.
    • divisor expressed as polynomial, e.g. standard CRC-16 : x¹⁶16+x¹⁵15+x²+1 (17 bit generator, yields 16 bit CRC)
    • if receiver gets non-zero remainder, error occurred.
  • encoding - codes can be assigned such that errors can be corrected as well as detected (not ASCII or unicode thoough)
    • Hamming distance: minimum # bits that must be flipped to change one valid code into another valid code.
    • Example: encoding 1001, 0011, 1100, 0110, 1010, 0101, 0000, 1111 has distance 2 and can encode 8 values.
    • n-bit errors can be detected if Hamming distance is n+1
    • n-bit errors can be corrected if Hamming distance is 2n+1
• Some detection techniques also permit correction - example above

Multiple (shared) access protocols

• Most LANs are broadcast, so hosts must share the medium
• Three strategies are employed
1. channel partitioning (e.g. TDM, FDM) - already covered
2. random access (allow for and recover from collision)
3. turn taking (e.g. token)
• for random access or turn taking, analogy to group conversation works

Random Access

• Station attempts transmission when packet received from network layer
• If two or more transmit at "about the same time", collision occurs
• Stations detect collision and back off a random amount of time before retransmitting
• Why random rather than fixed amount of time?
• CSMA with Collision Detection (CSMA/CD)
  • If collision detected before frame transmit complete, halt transmission immediately.
  • This can increase utilization
  • Will cover in more detail in conjunction with Ethernet (IEEE 802.3).

Turn Taking

• Polling protocols
  • Master/slave
  • Master polls each slave in turn
  • If slave has frame to transmit when polled, it transmits
  • Collision-free
  • Centralized control leaves it vulnerable
• Token-passing protocols
  • Token is “bathroom pass” that circulates on network.
  • Sender cannot transmit unless it has possession of token
  • Sender must release token at end of frame (or group of frames)
  • Collision-free
  • Token vulnerable
  • Token Ring (IEEE 802.5) is example

Addressing in the Link Layer

• Every network adapter (NIC) has a a link layer address
• Every frame must have link layer address as destination
• Link layer address commonly called MAC address
• MAC = Media Access Control
• We'll focus on Ethernet
• A host can have both IP and MAC address
  • Analogy: IP as postal address (can change), MAC as SS# (does not)
  • Host may need to work with network layers other than IP
  • Promotes independence between layers

MAC addresses

• Every Ethernet adapter card (NIC) has unique and "permanent" 48 bit MAC address
• Adapter manufacturers purchase chunks of the 48 bit address space from IEEE
• MAC address expressed as 12 hexadecimal digits: FF-FF-FF-FF-FF-FF

Address Resolution Protocol (ARP)

• ARP is Address Resolution Protocol. Primary use: IP-to-ethernet address translation
• Needed because the IP host is normally an ethernet station on a LAN
• ARP Message exchange protocol: Requester broadcasts request on the LAN; node whose IP address matches request generates response directly to requester (other nodes ignore request after verifying they are not target).

IEEE 802 project

• Collection of LAN standards
• Major sections are:
  • 802.1 : internetworking standards (network layer)
  • 802.2 : LLC specifications
  • 802.3 : Ethernet specifications
  • 802.4 : Token Bus specifications
  • 802.5 : Token Ring specifications
  • 802.11 : wireless LANs
• We will focus on IEEE 802.3, Ethernet

Ethernet Overview

• Originally developed at Xerox by Bob Metcalfe.
• Named after the "ether", a term describing the medium through which scientists once thought electromagnetic radiation flowed.
• Ether refers to the cable. Originally coax, now typically twisted pair or fiber optic
• Uses digital signal with Manchester encoding: each bit period has a signal transition in middle (low-to-high for 0, high-to-low for 1), to facilitate synchronization.
• Transmission rates have increased by orders of magnitude: originally 1Mbps (bits per second), then 10Mbps; Fast Ethernet is 100Mbps, Gigabit Ethernet is 1Gbps, current max about 100Gbps, with 1Tbps over fiber optic forecast by 2015.
• Note that M and G here are literally 1 million and 1 billion, not 2²⁰ and 2³⁰ as they are for storage capacity
• A number of standards exist. Examples include
  • 10Base-T : 10Mbps, baseband, Twisted pair cable
  • 100Base-T : 100Mbps, baseband, Twisted pair cable
  • 100Base-FX : 100Mbps, baseband, Fiber optic cable
  • 10GBase-T : 10Gbps, baseband, Twisted pair cable (2007)
  • Note: "baseband" means the signal uses entire bandwidth
• Characterized by bus topology and CSMA/CD for coordinating transmission.
• Connectionless! Unreliable! No ACKs!
• Reliability has to be implemented in higher layers (e.g. TCP) if desired

CSMA/CD : Frames Collide on the Common bus

• Consider how collisions can occur, since a station will not begin transmitting until it senses an idle carrier.
• If collision occurs, how to avoid repeating it, since both stations will try again? Solution is to have each station wait a different random amount of time before retrying.
• Specific Ethernet solution:
  • Think of "bit time" to be a small time slot - the time to transmit one bit
  • After i'th collision involving a frame, station waits random period based on i.
  • Range of random period is from 0 to 2ⁱ-1 bit times
  • This is called binary exponential backoff. Can you see why?
  • After first collision, they wait 0 or 1. With 2 stations, 50% probability of second collision.
  • After second collision, they wait 0 to 3. With 2 stations, 25% probability of third collision.
  • Etc. The maximum i is 10, the range 0 to 1023.
  • This is compromise, if period too short collisions reoccur; if too long bandwidth is wasted

Ethernet Addresses

• Ethernet uses 48-bit (6 byte) addresses.
• Recall that frames are broadcast.
  • NIC reads destination address as frame comes by.
  • If it matches the NIC i.d. number, the frame is saved and its payload passed on to the network layer of its station.
  • Otherwise, remainder of frame is ignored. Destination address is at the beginning of the frame.
• Some frames are truly broadcast, e.g. intended for all stations. A special destination address, all 1’s, is reserved for this purpose, and all NIC cards will match it.

Ethernet Frame

• Each frame contains both source and destination MAC address.
• Generic: consists of frame header part, with addressing and type information, and frame data part with the bits/bytes being transmitted (payload).

• field	bytes	description
  preamble	8	alternating 1 and 0 for synch (Manchester sine wave), last byte ends in "11"
  destination	6	all 1’s means broadcast, first bit 1 = group address, else individual
  source	6	physical address of source station
  type/length	2	#data bytes, or predefined frame type (mutually exclusive ranges)
  data	46-1500	If less than 46 bytes of data, is padded out to 46 bytes with zeroes. Necessary for minimum 512 bit frame length.
  CRC	4	Cyclic Redundancy Code

LAN extending device: Hub

• Ethernet defines maximum distance between stations
• Hubs are one means to extend
• Is a repeater: regenerates and retransmits input signals
• Physical layer device -- doesn't interpret bits it is carrying
• Multiple Ethernet stations plug into one hub to form star.
• Unless buffers provided, all stations must use same transmission protocols (e.g. 10Mbps)
• All stations connected through hub(s) comprise a collision domain, a.k.a. segment (different than Transport Layer segment!)

LAN extending device: Ethernet switch

• Conceptually like a router but has differences
  • Operates at only Physical and Link Layer (router also has Network layer)
  • Maintains forwarding table - Forwards based on MAC address (not IP address)
  • Has input interfaces and buffers, switching fabric, output interfaces and buffers
  • Forwards only within its LAN
• If only one station attached to each I/O interface then each station is a segment and no collisions occur!
• Ideal for connecting heterogeneous segments (e.g. 100Mbps, 1Gbps) because I/O interfaces have buffers - a.k.a bridging
• Can switch multiple transmissions at once, if each destination is different segment.
• Switch autonomously builds and maintains switch table, its forwarding table