1. Chapter 2 Multiple Access Protocols Professor Rick Han University of Colorado at Boulder rhan@cs.colorado.edu

2. Prof. Rick Han, University of Colorado at Boulder Announcements Previous lecture now online Homework #1 is on the Web site, due Feb. 5 Programming assignment #1 will be available on Web site on Tuesday, not today Next, Chapter 2, MAC protocols

3. Prof. Rick Han, University of Colorado at Boulder Recap of Previous Lecture Designing reliable ARQ protocols Acknowledgements Timeouts Sequence numbers Designing efficient reliable protocols Choose timeout wisely Keep the pipe full Stop-and-Wait: one outstanding packet Go-Back-N: sliding window, cumulative ACK Selective Repeat: sliding window, selective ACK

4. Prof. Rick Han, University of Colorado at Boulder Direct-Link or Point-to-Point Networks Physical layer handles bits Data-link layer handles packets Framing Error detection Retransmission-based protocols How do I send to N hosts? N point-to-point links Other possibilities?

5. Prof. Rick Han, University of Colorado at Boulder Shared-Media or Broadcast Networks N senders and receivers connected by a shared medium (copper wire, atmosphere-TV!) Sharing access to the same media Analogy: How do N persons converse in a room or at the dinner table? At once, or one by one? What is the communications protocol? Local Area Network (LAN) Ethernet (802.3) 802.11/Wireless Ethernet

6. Prof. Rick Han, University of Colorado at Boulder Multiple Access Protocols Determine which host is allowed to transmit next to a shared medium Channel reservation: TDMA, FDMA, CDMA, Token Ring, … Random access: ALOHA, CSMA/CD, CSMA/CA Ethernet 802.11/Wireless Ethernet

7. Prof. Rick Han, University of Colorado at Boulder Multiple Access Protocols (2) Also called Medium-Access Control (MAC) protocols Before data link-layer packets can be sent, a sender has to gain access to the media MAC layer is often placed in the stack between layer 2 and layer 1 Physical Layer MAC Layer Host A Data Link Layer

8. Prof. Rick Han, University of Colorado at Boulder Time Division Multiple Access (TDMA) Divide time into multiple slots Each host sends in a pre-determined slot Out-of-band reservation mechanism Compare to Time Division Multiplexing (TDM) …… 231123123 Host 1 Host 2 Host 3 1 2 3 Not Eth. Router/ Mux 1 2 12

9. Prof. Rick Han, University of Colorado at Boulder Frequency Division Multiple Access (FDMA) Divide spectrum into frequency bins Each host sends in a pre-determined frequency bin Out-of-band reservation mechanism Also called Frequency Division Multiplexing (FDM) Example: AM/FM radio, TV AM 500-1700 KHz FM 88-108 MHz Satellite GHz range Freq. (Hz) Host 1Host 2Host 3

10. Prof. Rick Han, University of Colorado at Boulder Code Division Multiple Access (CDMA) Use multiple orthogonal codes to partition a range of spectrum Each host sends using a pre-determined code Also called “spread spectrum” Direct-sequence spread spectrum (802.11, cell) Frequency-hopping spread spectrum (Bluetooth) Bluetooth F1F2F3 Hopping sequence: F1, F3, F2, F1, F3, F2, … Host 1’s Code: 132, Host 2’s Code: 321, Host 3’s Code: 213 – all 3 codes orthogonal Host 1 Host 2 Host 3 Freq:

11. Prof. Rick Han, University of Colorado at Boulder Random Access: ALOHA Protocol Developed at University of Hawaii in 1971 by Abramson Ground-based UHF radios connect computers on several island campuses to main university computer on Oahu “pure” ALOHA: hosts transmit whenever they have information to send – form of random access Collision will occur when two hosts try to transmit packets at the same time Hosts wait a timeout=1 RTT for an ACK. If no ACK by timeout, then wait a randomly selected delay to avoid repeated collisions, then retransmit

12. Prof. Rick Han, University of Colorado at Boulder Random Access: ALOHA Protocol (2) Collision of packets can occur when a packet overlaps another packet Packet A Packet CPacket B time T0 Collision Wasted Time Colliding with B Wasted Time Due to a Collision = 2 packet intervals

13. Prof. Rick Han, University of Colorado at Boulder Random Access: Slotted ALOHA Rather than sending a packet at any time, send along time slot boundaries Collision are confined to one time slot Packet A Packet CPacket B time T0 Collision No Collision Wasted Time Due to a Collision = 1 packet interval

14. Prof. Rick Han, University of Colorado at Boulder Random Access: Slotted ALOHA (2) How do hosts synchronize to begin transmitting along time slot boundaries? One central station transmits a synchronization pulse or beacon Slotted ALOHA is more efficient than ALOHA because when there is a collision, the wasted time is confined to one time slot Assuming Poisson packet arrivals (memoryless), can compute the maximum throughput of ALOHA to be 18%. Maximum throughput of Slotted ALOHA is 37% Why are ALOHA & slotted ALOHA so inefficient?

15. Prof. Rick Han, University of Colorado at Boulder Random Access: CSMA ALOHA & slotted ALOHA are inefficient because hosts don’t take into account what other hosts are doing before they transmit Example: at party, everyone speaks whenever they want to, regardless of whether another person is speaking Instead, “listen before you talk” = Carrier Sense Multiple Access (CSMA) Sense for “carriers” (see if anyone else is transmitting) before you begin transmitting Packet A time Host B listens Packet BPacket YPacket X Host B sends delay Collision still possible over long prop. delays

16. Prof. Rick Han, University of Colorado at Boulder Random Access: 1-Persistent CSMA If channel is busy, A host listens continuously When channel becomes free, a host transmits its packet immediately (with probability 1) Collision scenarios Hosts A and B are far apart (long prop. delay). A’s signal takes a long time to reach B. So, B thinks channel is free, and begins transmitting. Hosts B and C transmit as soon as A finishes Still, CSMA is more efficient than ALOHA variants Packet A time Host B listens Packet B Packet YPacket X Host B sendsCollision

17. Prof. Rick Han, University of Colorado at Boulder Random Access: p-Persistent CSMA Generalization of 1-persistent CSMA Typically applied to slotted channels Slot length is chosen as maximum propagation delay A host senses the channel, and If slot is idle, transmit with probability p, or defer with probability q=1-p If next slot is idle, transmit with probability p, or defer with probability 1-p, repeat… If channel is busy, then sense channel continuously until it becomes free, begin again

18. Prof. Rick Han, University of Colorado at Boulder Random Access: Non-Persistent CSMA Host does not sense channel continuously Instead, if channel is busy, Wait/sleep a random interval before sensing again As with 1-persistent CSMA, as soon as channel is idle, then send a packet Random interval reduces collisions Higher throughput than 1-persistent CSMA when many senders Packet A time Host B listens Packet B Host B sends Random Sleep

19. Prof. Rick Han, University of Colorado at Boulder Random Access: Ethernet CSMA/CD Ethernet uses CSMA/CD, i.e. CSMA with Collision Detection (CD) “Listen-while-talk” protocol A host listens even while it is transmitting, and if a collision is detected, stops transmitting Packet A time Host B senses carrier Packet B delay Host B starts sending Packet B Host B detects collision And stops sending Not transmitted

20. Prof. Rick Han, University of Colorado at Boulder Random Access: Ethernet CSMA/CD (2) Can abort transmission sooner than end-of- packet if there is a collision Can happen if prop. delays are long Better efficiency than pure CSMA CSMA/CD doesn’t require explicit acknowledgement Unlike CSMA, which requires an ACK or timeout to detect a collision Collision detection is built into the transmitter When collision detected, begin retransmission

21. Prof. Rick Han, University of Colorado at Boulder Random Access: Ethernet CSMA/CD (3) Exponential backoff strategy When a collision is detected, a host waits for some randomly chosen time, then retransmits a packet If a second collision is detected, a host doubles the original wait time, then retransmits the packet Each time there is another collision, the wait time is doubled before retransmission Variants: At each retransmission, choose a random value from the exponentially increasing wait time. At each retransmission, choose randomly from among a discrete set of values within exponentially increasing wait time Retransmit a finite # of times

22. Prof. Rick Han, University of Colorado at Boulder Random Access: Ethernet CSMA/CD (4) CSMA/CD can be used with nonpersistent, 1- persistent, or p-persistent variants of CSMA Ethernet is synonymous with the IEEE 802.3 standard Initial work on Ethernet at Xerox in early 70’s Ethernet specifies 1-persistent CSMA/CD To extend an Ethernet, repeaters are placed. Start to run into propagation delay issues and noise amplification issues Ethernet keeps its maximum length to 2500 m to keep prop. delays tight, so that CSMA/CD responds well

23. Prof. Rick Han, University of Colorado at Boulder Random Access: 802.11 “Wireless” Ethernet Employs CSMA/CA, i.e. CSMA with Collision Avoidance (CA) Hidden terminal effect Example: B can hear A and C, but A and C can’t hear each other. If A is sending B, C thinks channel is clear and starts sending => collision! Doesn’t happen in wired Ethernet, because hosts can hear each other Host A Collision Host BHost C

24. Prof. Rick Han, University of Colorado at Boulder Random Access: 802.11 “Wireless” Ethernet (2) How to handle the hidden terminal effect? Host A sends a Request-To-Send (RTS) Host B sends a Clear-To-Send (CTS) Host C hears the CTS, and does not interrupt transmission between A and B This helps implement Collision Avoidance Host A Host C Suppresses Its Data Host BHost C RTS CTS Data

25. Prof. Rick Han, University of Colorado at Boulder Random Access: 802.11 “Wireless” Ethernet (3) Acknowledgements are still needed After Host A has finished sending its data, Host B sends an ACK Host C hears this ACK, and sends its RTS Host AHost BHost C Data ACK RTS

26. Prof. Rick Han, University of Colorado at Boulder Random Access: 802.11 “Wireless” Ethernet (4) 802.11 does not support Collision Detection In a wired LAN, transmitter can check voltage levels to see if there is a collision. A remote transmitter’s power doesn’t attenuate severely. In a wireless LAN, the transmitter’s power overwhelms a distant transmitter’s power, so it’s difficult to detect collision. What happens if two RTS’s collide? Senders realize after timeout that CTS did not come back, and then practice exponential backoff in trying to send new RTS’s