nnouncements CS 5565 Network rchitecture and Protocols Lecture 4 odmar ack Project due in parts: pr 5 and May xtra Credit Opportunities: xpand simulator (and your implementation) to introduce multiple link failures and link resurrection thernet thernet dominant wired LN technology: cheap <$0 for 00Mbps! first widely used LN technology standardized in 80.3 series of I standards simpler, cheaper than token LNs and TM kept up with speed race: 0 Mbps 0 bps Metcalfe s thernet sketch 4 thernet Model Connectionless: No handshaking between sending and receiving adapter Unreliable: No acks (or nacks) higher-level protocols must compensate, but low R (bit error rate) CSM/CD: no slots random access protocol 5 Ideal Multiple ccess Protocol roadcast channel of rate R bps. When one node wants to transmit, it can send at rate R.. When M nodes want to transmit, each can send at average rate R/M 3. Fully decentralized: no special node to coordinate transmissions no synchronization of clocks, slots 4. Simple 6
MC Protocols: a taxonomy MC Media ccess Control Three broad classes: Channel Partitioning divide channel into smaller pieces (time slots, frequency, code) allocate piece to node for exclusive use Random ccess channel not divided, allow collisions recover from collisions Taking turns Nodes take turns, but nodes with more to send can take longer turns Star Topology us topology popular through mid 90s Now star topology prevails Connection choices: or Hubs rarely used nowadays or 7 8 thernet Frame Structure Sending adapter encapsulates IP datagram (or other network layer protocol packet) in thernet frame 7 6 6 ----------- 4 <05DC 80.3 Length 0800 IP 0806 RP 809 ppletalk Preamble: 7 bytes with pattern 0000 followed by one byte with pattern 000 used to synchronize receiver, sender clock rates 9 thernet CSM/CD algorithm. daptor receives datagram from net layer & creates frame. If adapter senses channel idle, it starts to transmit frame. If it senses channel busy, waits until channel idle and then transmits 3. If adapter transmits entire frame without detecting another transmission, the adapter is done with frame! 4. If adapter detects another transmission while transmitting, aborts and sends jam signal 5. fter aborting, adapter enters exponential backoff: after the mth collision, adapter chooses a K at random from {0,,,, m -}. dapter waits K 5 bit times and returns to Step 0 thernet s CSM/CD (more) Jam Signal: make sure all other transmitters are aware of collision; 48 bits it time:. microsec for 0 Mbps thernet ; for K=03, wait time is about 50 msec See/interact with Java applet on WL Web site: highly recommended! xponential ackoff: oal: adapt retransmission attempts to estimated current load heavy load: random wait will be longer first collision: choose K from {0,}; delay is K 5 bit transmission times after second collision: choose K from {0,,,3} after ten collisions, choose K from {0,,,3,4,,03} CSM/CD efficiency t prop = max prop between nodes in LN t trans = time to transmit max-size frame efficiency 5t / prop t trans fficiency goes to as t prop goes to 0 oes to as t trans goes to infinity Much better than LOH, but still decentralized, simple, and cheap
fficiency of 0Mbps thernet thernet Physical Layer ackoff slot times is 5-bit N: does not imply CSM/CD thernet is slotted 3 4 Manchester ncoding Used in 0aseT ach bit has a transition llows clocks in sending and receiving nodes to synchronize with each other no need for a centralized, global clock among nodes! 5 Fast thernet: 00aseT 45 it ncoding MLT-3 (Multi-Level Threshold) Source: http://de.wikipedia.org 0000 0 000 000 000 000 00 00 000 000 00 00 00 00 0 0 000 000 00 00 00 00 0 0 00 00 0 0 0 00 0 6 igabit thernet uses standard thernet frame format allows for point-to-point links and shared broadcast channels in shared mode, CSM/CD is used; short distances between nodes required for efficiency supports s, called here uffered Distributors Though you wouldn t use them full-duplex at bps for point-to-point links 000aseT: 80 coding with 5 signal levels See IOL page for more information Hubs Hubs are essentially physical-layer repeaters: bits coming from one link go out all other links at the same rate no frame buffering no CSM/CD at : adapters detect collisions provides net management functionality Single collision domain 7 8 3
Interconnecting LNs Hypothetical design: backbone interconnects LN segments Disadvantage: individual segment collision domains become one large collision domain can t interconnect 0aseT & 00aseT Switches Link layer device stores and forwards thernet frames examines frame header and selectively forwards frame based on MC dest address when frame is to be forwarded on segment, uses CSM/CD to access segment transparent hosts are unaware of presence of es plug-and-play, self-learning es do not need to be configured 9 0 Forwarding 3 How do determine onto which LN segment to forward frame? Looks like a routing problem... N: Switches are called bridges if #interfaces == has a table Self Learning entry in table: (MC ddress, Interface, Time Stamp) stale entries in table dropped (TTL can be 60 min) learns which hosts can be reached through which interface when frame received, learns location of sender: incoming LN segment records sender/location pair in table Switch xample Suppose C sends frame to D Switch xample () Suppose D replies back with frame to C. 3 address interface I D F C H Switch receives frame from C notes in bridge table that C is on interface because D is not in table, forwards frame into interfaces and 3 frame received by D 3 3 C D Switch receives frame from from D notes in bridge table that D is on interface because C is in table, forwards frame only to interface frame received by C F 4 H address interface I C 3 4
Switch: Traffic Isolation installation breaks subnet into LN segments filters packets: same-ln-segment frames not usually forwarded onto other LN segments segments become separate collision domains Nowadays: direct to host connection -> no collisions! collision domain collision domain collision domain 5 Switches: Dedicated ccess Switch with many interfaces Hosts have direct connection to No collisions; full duplex -to- and -to-f simultaneously cut-through ing: frame forwarded from input to output port without first collecting entire frame slight reduction in latency C F 6 D This is the dominant technology today Switches: Flow Control What if =00Mbps, =0Mbps? host sends frames faster than can send them out of outgoing port undetected losses due to overflow could occur I 80.3 provides LLC layer that has flow control facilities pause frames to get sender to stop to external network Switches vs Routers router mail server web server IP subnet 7 8 Spanning Tree ridges Switches self-configure into spanning tree 9 80.D ased on [Perlman 985] Spanning Tree algorithm Root selection by lowest MC ID Link-State approach: HLLO messages Links are in states: CKUP (inactive) or FORWRDIN (active) and in-between states reaks loops fast, recovery takes up to min Disadvantage: Slow convergence after breakage 80.w: RSTP (Rapid STP) Lack of multipath capability Not aware of VLNs (discussed next) 30 5
Virtual LNs Virtual LNs () (a) Four physical LNs organized into two VLNs, gray and white, by two bridges. (b) The same 5 machines organized into two VLNs by es. 3 3 The I 80.Q Standard Transition from legacy thernet to VLNaware thernet. The shaded symbols are VLN aware. The empty ones are not. The I 80.Q Standard () 80.Q thernet frame format VLN ssociation may be done Via port Via MC address Via Layer 3 information (IP address) 33 34 Switches vs. Routers both store-and-forward devices routers: network layer devices (examine network layer headers) es are link layer devices routers maintain routing tables, implement routing algorithms es maintain tables, implement filtering, learning algorithms + spanning tree algorithms Hubs vs. Routers vs. Switches Traffic isolation Hubs Routers Switches No Yes Yes Plug & Play Yes No Yes Optimal Routing Cut Through No Yes No Yes No Yes 35 36 6