1. cs/ee 143 Communication Networks Chapter 3 Ethernet Text: Walrand & Parakh, 2010 Steven Low CMS, EE, Caltech

2. Warning These notes are not self-contained, probably not understandable, unless you also were in the lecture They are supplement to not replacement for class attendance

3. Agenda Ethernet history/devices Switch Ethernet forwarding table Spanning tree protocol Little’s theorem (informal proof)

4. Ethernet Each (layer 2) network full connectivity: every node can reach every other node broadcast capable: every node (inc. router) can broadcast to all other nodes e.g. Ethernet, WiFi, cable network, etc.

5. Aloha network (1970) Randomized multiple access Send on frequency f1; receive ack on frquency f2. If no ack after timeout, wait a random time and re-transmit

6. Aloha network (1970) Randomized multiple access If an ack is not outstanding, transmit immediately If no ack, re-transmit after a random delay

7. Aloha network (1970) Randomized multiple access Max utilization (prob of success) ~ 1/e ~ 37%

8. Slotted Aloha utilization Model Slotted time, fixed packet size, n stations 1 slot = 1 pkt transmission time In each slot, each station transmits independently with probability p Prob (slot t has a successful transmission)

9. Slotted Aloha utilization

12. Unslotted Aloha utilization Model Fixed packet size, n stations Slotted time of duration  << 1. pkt transmission time = 1/  In each  slot, each station transmits independently with probability p Prob (slot  has a successful transmission)

13. Unslotted ALOHA utilization Prob (a pkt transmission started in an arbitrary  -slot by station 1 is successful)

14. Unslotted ALOHA utilization Prob (a pkt transmission started in an arbitrary  -slot by station 1 is successful)

15. Unslotted ALOHA utilization

20. Ethernet cable (1973-76) CSMA/CD (carrier sensing multiple access/collision detection) 1.Wait till channel is idle; wait for a random time. 2.Transmit when the channel is idle following the random wait. 3.Abort if collision is detected, and goto 1.

21. Ethernet cable (1973-76) Truncated binary exponential backoff 1.Pick X uniformly at random from {0, 1,..., 2^n-1}, n = min (10, m), m = #collisions. Give up & declare error when m = 16. 2.Wait X x 512 bit times (4,096 bits for 1G) 3.If collide, increment m and repeat.

22. Ethernet cable (1973-76) Capture or winner-takes-all effect A station that collides is more likely to pick a larger random backoff time. A station that successfully transmits is more likely to pick a smaller backoff time and hence more likely to successfully transmit again

23. Ethernet hub (1980s) CSMA/CD as in Ethernet cable

24. Ethernet hub (1980s) A station waits a random multiples of T = 2 PROP before transmitting When n stations transmit independently with prob p, then prob of success is <= 1/e when n is large Hence avg time till first success = e T Utilization = TRANS / (TRANS + (e-1)T) = 1 / (1 + 3.4A), A = PROP/TRAN

25. Ethernet switch Ethernet switch eliminates collision, provided switch buffer is big enough.

26. Ethernet switch: forwarding table (Ethernet) MAC address 1.48 bit 2.Globally unique to MAC device, location independent (c.f. IP) 3.Broadcast address: 48 ones

27. Ethernet switch: forwarding table x  y: [ y | x | data ]

28. Agenda Ethernet history/devices Switch Ethernet forwarding table Spanning tree protocol Little’s theorem (informal proof)

29. Ethernet switch routing: STP Goal Operation Example Performance x  y: [ y | x | data ]

30. Spanning tree protocol Goal: for all switches in a LAN to compute a shortest-path tree used to route layer-2 packets one tree for entire LAN rooted at the switch with the smallest ID (MAC address) decentralized, asynchronous, robust computation

31. Spanning tree protocol Three criteria 1.The root switch always forwards messages on all its ports 2.Each switch computes its shortest path (in #bridges) to root 3.All switches connected to a LAN elect a designated switch for the LAN to send packets towards root switch  A switch that is not elected for any of the LANs it is connected to will not receive nor forward any data packet

32. Spanning tree protocol Switches send packets asynchronously with [ my ID | current root ID | distance to root ] A switch relays packets whose “current root ID” is the smallest it has seen so far (& smaller than its own “current root ID”), and adds 1 to “distance to root” If the “distances to root” on STP packets received by a switch on all its ports are the same or smaller than what it believes its distance is, then the switch stops forwarding … until protocol converges Completely decentralized, asynchronous, robust

33. STP: example I’m 3 I think root is 3 my distance to root is 0

34. STP: example I’m 3 I think root is 3 my distance to root is 0

35. STP: example a new initiation before previous converges

36. STP: example a new initiation before previous converges

37. STP: example a new initiation before previous converges

38. STP: example During transient, B5 may connect to root B1 either via B3 or B4 – which should B5 use? Ans: use switch with a smaller ID (B3)

39. Spanning tree for all switches x  y: [ y | x | data ]

40. STP: designated switches B4 believes its distance to root B1 is 2 The STP packets from both its ports have distances equal or less. So it does not forward and is not a designated switch for neither LAN Neither B4 nor B5 will be involved in forwarding data packets

41. Spanning tree protocol Performance Unique path between every source- destination path Can potentially be bad since 2 switches may communicate only via root  e.g. in a ring of switches, the switch with the largest ID communicates with root via the longest path Penalty is usually not too bad since it is in a LAN

42. Agenda Ethernet history/devices Switch Ethernet forwarding table Spanning tree protocol Little’s theorem (informal proof)

43. Little’s law

50. Queueing system random arrival process with rate random service time with average  Little’s law  Verifies directly for M/M/1, but holds much more generally  Extremely useful because of its generality

51. M/M/1 queue Poisson arrival process with rate Exponential service time with average

52. Queueing system random arrival process with rate random service time with average