CS 5565 Network Architecture and Protocols Lecture 3 Godmar Back

Announcements Assignment: Create CS5565 Forum Do Wireshark Lab 1 Don’t need to hand it in. Create CS5565 Forum Use this to find a project partner All 4 projects will be done in groups of up to 2. CS 5565 Spring 2009 5/26/2018

Summary Terminology: hosts (end systems), communication links, routers, transmission rates, packets, internet vs. intranet vs. the Internet Protocols: protocols define format, order of messages sent and received among network entities, and actions taken on msg transmission, receipt View from network edge: Client/server, peer2peer, other models Service view Communication infrastructure provide connection-oriented + connectionless service View from network core: Circuit-switching vs packet-switching Datagram network vs. virtual-circuit networks CS 5565 Spring 2009 5/26/2018

How do loss and delay occur? packets queue in router buffers packet arrival rate to link exceeds output link capacity packets queue, wait for turn B A packet being transmitted (delay) packets queueing (delay) free (available) buffers: arriving packets dropped (loss) if no free buffers CS 5565 Spring 2009 5/26/2018

Four sources of packet delay 1. Nodal processing delay: check bit errors determine output link 2. Queuing delay time waiting at output link for transmission depends on congestion level of router A B propagation transmission nodal processing queueing CS 5565 Spring 2009 5/26/2018

Queuing Delay Show Applet here Queuing delay depends on http://media.pearsoncmg.com/aw/aw_kurose_network_2/applets/queuing/queuing.html Queuing delay depends on traffic intensity nature of packet arrival process (bursts, periodic, periodic bursts, random intervals) CS 5565 Spring 2009 5/26/2018

Queueing Delay traffic intensity = La/R R=link bandwidth (bps) L=packet length (bits) a=average packet arrival rate traffic intensity = La/R La/R ~ 0: average queueing delay small La/R  1: delays become large La/R > 1: more “work” arriving than can be serviced, average delay infinite! CS 5565 Spring 2009 5/26/2018

Queuing Analysis Source: Ts : mean service time Stallings Ts : mean service time Coefficient of variation determines increase in delay CS 5565 Spring 2009 5/26/2018

Queuing Analysis (II) Depending on estimated coefficient of variance, pick appropriate server model M/M/1: both arrival rate & service time is “M”, Poisson process vs. negative exponentially distributed services In general: X/Y/n, where n number of servers & X, Y one of G: general independent M: negative exponential distribution D: deterministic (fixed length of service) Q.: what’s the expected queue length for a D/D/1 queue if arrival rate < service rate? CS 5565 Spring 2009 5/26/2018

Delay in packet-switched networks 3. Transmission delay: R=link bandwidth (bps) L=packet length (bits) time to send bits into link = L/R 4. Propagation delay: d = length of physical link s = propagation speed in medium (~2x108 m/sec) propagation delay = d/s A B propagation transmission nodal processing queueing Note: s and R are very different quantities! CS 5565 Spring 2009 5/26/2018

Transmission vs. Propagation Delay Show Applet here http://media.pearsoncmg.com/aw/aw_kurose_network_2/applets/transmission/delay.html Transmission delay depends on speed of link (10Mbps, 1000Mbps, …) Propagation delay depends on distance (and speed of light in medium) CS 5565 Spring 2009 5/26/2018

Nodal delay dproc = processing delay dqueue = queuing delay typically a few microsecs or less dqueue = queuing delay depends on congestion dtrans = transmission delay = L/R, significant for low-speed links dprop = propagation delay a few microsecs to hundreds of msecs CS 5565 Spring 2009 5/26/2018

Packet-switching: store-and-forward L R R R Takes L/R seconds to transmit (push out) packet of L bits on to link or R bps Entire packet must arrive at router before it can be transmitted on next link: store and forward Example: L = 7.5 Mbits R = 1.5 Mbps delay = ? (assume no propagation/processing/queuing delay) 15 seconds CS 5565 Spring 2009 5/26/2018

End-to-end vs. nodal delay Question: in store-and-forward model, does end-to-end delay for a message of length L depend on the number/size of packets the message is split into? Let’s look at: http://media.pearsoncmg.com/aw/aw_kurose_network_2/applets/message/messagesegmentation.html CS 5565 Spring 2009 5/26/2018

End-to-end vs. nodal delay Question: in store-and-forward model, end-to-end delay for a message of length L depends on the number/size of packets the message is split into: Transmission times of packets can be overlayed if multiple packets are part of a message “Store-and-forward” model applies to packet, not message Consequence: packetization reduces transmission delay But you pay a price for header overhead CS 5565 Spring 2009 5/26/2018

Circuit vs. Packet Switching (2) Circuit Switching Dedicated link bandwidth Dedicated switch capacity Low link utilization Low overall utilization Bounded delay variance Packet Switching Better Link utilization Better overall utilization Need for congestion control Need to identify to which “call” a packet belongs Smaller average delay than TDM High variance in delay CS 5565 Spring 2009 5/26/2018

Latency vs Bandwidth Latency (Delay) lags Bandwidth [Patterson 2004] Similar pattern in many areas CS 5565 Spring 2009 5/26/2018

Bandwidth Delay Product Delay d Bandwidth b Aka “size of pipe” Important in protocol design b in bps d in ms bxd in bytes Gigabit Ethernet 1G .150 18,750 100Mbit Ethernet 100M .7 8,750 Dialup Modem+Internet 56K 180 1,260 Cable Modem+Internet 4M 67,500 CS 5565 Spring 2009 5/26/2018

traceroute Provides delay measurement from source to router along end-end Internet path towards destination Problem: Don’t know which route is taken How: Send probes to destination Tell probes to die off after i hops, i = 1..30 Ask router to send echo packets if packets dies Measure RTT 3 probes 3 probes 3 probes CS 5565 Spring 2009 5/26/2018

Example Internet delays and routes traceroute: from host in Silicon Valley (keeda.stanford.edu) to host in Frankfurt, Germany (www.titanic-magazin.de) > traceroute www.titanic-magazin.de traceroute to www.titanic-magazin.de (62.75.228.90), 30 hops max, 38 byte packets 1 Gates-rtr.Stanford.EDU (171.64.72.1) 0.523 ms 0.339 ms 0.304 ms 2 bbr2-rtr.Stanford.EDU (171.64.1.161) 0.401 ms 0.346 ms 0.334 ms 3 border2-rtr.Stanford.EDU (171.64.1.148) 4.288 ms 1.070 ms 1.458 ms 4 g1.ba21.b003123-1.sfo01.atlas.cogentco.com (66.250.7.137) 5.231 ms 7.975 ms 9.097 ms 5 g1-1.core02.sfo01.atlas.cogentco.com (66.28.6.13) 11.364 ms 16.192 ms 16.961 ms 6 p14-0.core01.dca01.atlas.cogentco.com (66.28.4.210) 85.497 ms 84.084 ms 80.291 ms 7 p2-0.core01.iad01.atlas.cogentco.com (154.54.2.202) 89.268 ms 88.548 ms 90.046 ms 8 lambdanet.iad01.atlas.cogentco.com (154.54.11.162) 156.812 ms 200.935 ms 157.819 ms 9 LON-2-pos210.uk.lambdanet.net (81.209.156.29) 159.647 ms 159.709 ms 166.504 ms 10 DUS-2-pos700-0.de.lambdanet.net (82.197.136.18) 176.365 ms 163.668 ms 165.177 ms 11 DUS1-5029.de.lambdanet.net (217.71.104.30) 171.229 ms 173.782 ms 171.486 ms 12 titanic.luka.de (62.75.228.90) 172.654 ms 183.307 ms 173.239 ms CS 5565 Spring 2009 5/26/2018

“Real” Internet delays and routes traceroute: from host in Silicon Valley (keeda.stanford.edu) to host in Frankfurt, Germany (www.titanic-magazin.de) > traceroute www.titanic-magazin.de traceroute to www.titanic-magazin.de (62.75.228.90), 30 hops max, 38 byte packets 1 Gates-rtr.Stanford.EDU (171.64.72.1) 0.523 ms 0.339 ms 0.304 ms 2 bbr2-rtr.Stanford.EDU (171.64.1.161) 0.401 ms 0.346 ms 0.334 ms 3 border2-rtr.Stanford.EDU (171.64.1.148) 4.288 ms 1.070 ms 1.458 ms 4 g1.ba21.b003123-1.sfo01.atlas.cogentco.com (66.250.7.137) 5.231 ms 7.975 ms 9.097 ms 5 g1-1.core02.sfo01.atlas.cogentco.com (66.28.6.13) 11.364 ms 16.192 ms 16.961 ms 6 p14-0.core01.dca01.atlas.cogentco.com (66.28.4.210) 85.497 ms 84.084 ms 80.291 ms 7 p2-0.core01.iad01.atlas.cogentco.com (154.54.2.202) 89.268 ms 88.548 ms 90.046 ms 8 lambdanet.iad01.atlas.cogentco.com (154.54.11.162) 156.812 ms 200.935 ms 157.819 ms 9 LON-2-pos210.uk.lambdanet.net (81.209.156.29) 159.647 ms 159.709 ms 166.504 ms 10 DUS-2-pos700-0.de.lambdanet.net (82.197.136.18) 176.365 ms 163.668 ms 165.177 ms 11 DUS1-5029.de.lambdanet.net (217.71.104.30) 171.229 ms 173.782 ms 171.486 ms 12 titanic.luka.de (62.75.228.90) 172.654 ms 183.307 ms 173.239 ms CS 5565 Spring 2009 5/26/2018

Routing across Tiers Tier 1 ISP NAP local Tier 3 ISP Tier-2 ISP CS 5565 Spring 2009 5/26/2018

Tiers of ISP (cont’d) Tier 1: usually carriers, own networks/fiber, interconnect with Tier 1 ISPs on reciprocal basis (8 interconnection regions in US), examples: UUnet, Level3, Sprint, CW … Tier 2: national/regional, connects to Tier 1 (“buys transit”), but peer among each other (AOL, Adelphia, Comcast) - driven by P2P traffic Tier 3: regional/local providers Definitions fluent, some are only Tier 1 in some regions Terms: POP (point of presence) NAP (network access point) - public interchanges (Equinix) Private peering points/links CS 5565 Spring 2009 5/26/2018

Changing Landscape of Peering Before 2000 Crash Source: [Norton 2004] Now LSNSC (large scale network savvy content providers) CS 5565 Spring 2009 5/26/2018

Aside: Estimating Bottleneck Bandwidth along a Path Network core doesn’t provide this information “packet-pair” method (van Jacobson) Send data in back-to-back packets Difference between acks is related to delay experienced along slowest link Many other methods developed, big research area see for example [M Goutelle 2003] CS 5565 Spring 2009 5/26/2018

Summary Transmission & Propagation Delay End-to-end delay in packet-switched networks Traceroute + network diagnostics Structure of Internet Bandwidth-delay product CS 5565 Spring 2009 5/26/2018