Transcript ppt
ECEN5553 Telecom Systems Dr. George Scheets Week #8 Readings: [14a] "Can You Trust Your Fridge" [14b] "How a Teakettle Can Kill Your Cloud" [15a] "Hacking the Voter" [15b] "Hacking a Voting Machine" [16] "Voice Over the Internet: A Tutorial" Exam 1 Final Results (90 points max) Hi = 88.4, Low = 37.2, Average = 70.80, Deviation = 11.99 A > 80, B > 67, C > 58, D > 49 Outline: Lecture 22, 5 October (Live) No later than 12 October (Remote DL) Exam #2: 24 October (Live & Local DL) No Later than 31 October (Remote DL) Outlines Received due 5 October (local) 12 October (remote) 64 % Carrier Ethernet Status 2009 U.S. Market Revenue $1.5 Billion 2010 $3.2 Billion 2013 $5.5 Billion 2016 $11.1 Billion (projected) 2018 $13 Billion (projected) Backhaul from wireless cell sites a major growth area source: www.accedian.com www.telecompetitor.com MAN/WAN Connectivity Options Carrier Ethernet Carrier Switches are Ethernet frame aware PBB I/O decisions based on Layer 2 Ethernet Address IP/MPLS I/O decisions based on MPLS tag Virtual Circuits can be used StatMux BW required based more so on average input rates Pricing function of peak rate, CIR, priority, and maybe distance On the way in. 21st century version of Frame Relay Carrying Capacity Line Speed Active Idle Application Traffic Overhead Carrying Capacity = Traffic(bps)/Line Speed(bps) Goodput = Application Traffic Carried (bps) Queue Length 100,000,000 bps output trunk 100,000,001 bps average input Average Input rate > Output rate Queue Length builds up (without bound, in theory) Queue Length 100,000,000 bps output trunk 99,999,999 bps average input Average Input rate < Output rate Queue Length not infinite... ...but very large Queue Length @ 100% Load Output capacity = 7 units Input = 7 units on average (two dice rolled) t1: input = 4, output = 4, queue = 0 t2: input = 5, output = 5, queue = 0 t3: input = 4, output = 4, queue = 0 t4: input = 7, output = 7, queue = 0 t5: input = 11, output = 7, queue = 4 t6: input = 10, output = 7, queue = 7 t7: input = 6, output = 7, queue = 6 t8: input = 5, output = 7, queue = 4 t9: input = 8, output = 7, queue = 5 t10: input = 11, output = 7, queue = 9 This queue will tend to get very large over time. Queue Length @100% Load Will tend to increase w/o Bound. 34000 3.40910 queue5 j2000 0 0 0 0 2 10 4 10 5 5 6 10 8 10 5 j 5 1 10 6 110 5 6 32000 1.98310 queue5 j1000 0 0 0 0 2 10 5 4 10 5 6 10 5 j 5 8 10 5 1 10 6 110 6 "Die Roll" Queue Lengths 101% Load 34000 3.40910 100% Load queue5 j2000 99% Load, Average Queue = 44.46 0 0 0 0 2 10 5 4 10 5 6 10 5 j 5 8 10 5 1 10 6 110 6 Real vs Artificial Trace 10 Seconds Real Traffic 10 Seconds Artificial M/M/1 Traffic Source: Willinger et al, "Self-Similarity through High Variability", IEEE/ACM Transactions on Networking, February 1997. Real vs Artificial Trace 100 Seconds Real Traffic 100 Seconds Artificial M/M/1 Traffic Real vs Artificial Trace 16.7 Minutes Real Traffic 16.7 Minutes Artificial M/M/1 Traffic Real vs Artificial Trace 167 Minutes Real Traffic 167 Minutes Artificial M/M/1 Traffic Real vs Artificial Trace 27.78 Hours Real Traffic 27.78 Hours Artificial M/M/1 Traffic Self Similar Behavior Infinite Length Queue (Classical StatMux Theory) Probability of dropped packets Average Delay for delivered packets 0% Trunk Offered Load 100% Finite Length Queue (Real World StatMux) Probability of dropped packets Average Delay for delivered packets 0% Trunk Offered Load 100% You could fully load StatMux trunk lines... but your customers would be screaming at you due to lousy service. Switched Network Carrying Capacity Line Speed: Traffic injection speed Efficiency: Ability to use that Line Speed Throughput: bps of traffic (+ overhead) moved = Efficiency * Line Speed Carrying Capacity: Ability to usefully use Line Speed Accounts for packet overhead Accounts for inability to fully load trunk lines with StatMux'd traffic & still have a usable connection Goodput: bps of application traffic moved = Carrying Capacity * Line Speed Carrying Capacity Line Speed Active Traffic Idle Overhead Carrying Capacity = (%Trunk Load) * (%Traffic) = Traffic(bps)/Line Speed(bps) Packet Switch StatMux Trunking Pure Internet (or Ethernet) Model Fixed Rate Traffic Bursty Data Traffic Router SONET & OTN (Ethernet) Assumptions: All Fixed Rate Traffic is packetized. All traffic is Statistically Multiplexed onto the trunk BW. Internet Service Provider Backbone Packet Aware Router StatMux, Packet Switched Network, Full Duplex Trunks. Access lines mostly attach to routers. ATM Trunking (In Nineties, claimed as Tomorrow's Network Model) Fixed Rate Traffic Bursty Data Traffic ATM Switch SONET OC-N Assumptions: Fixed Rate Traffic gets CBR Virtual Circuits. CBR traffic gets near-TDM like service. Data Traffic is StatMuxed onto the remaining trunk BW. ATM Backbone Cell Aware ATM Switch StatMux/TDM, Cell Switched Network, Full Duplex Trunks. Access lines mostly attached to ATM switches, and "ATM capable" routers, FR switches, TD Muxes, & cross connects. Circuit Switch TDM Trunking (Eighties 'Private Line' Network Model) Fixed Rate Traffic TDM Switch Bursty Data Traffic Fiber, Cable, & Microwave Assumptions: All Traffic receives trunk bandwidth based on peak input rates. No aggregation. Data traffic consists of many slower speed, relatively lightly loaded circuits. Carrier Leased Line Backbone Byte Aware Cross-Connect TDM, Circuit Switched Network, Full Duplex Trunks. Access lines mostly attach to routers, FR & ATM switches, TD Muxes, & cross connects of other carriers. Hybrid TDM Trunking (Network Model for older Carriers) Fixed Rate Bursty Data Packet Switch TDM Switch SONET Assumptions: Bursty Data Traffic is all StatMuxed onto a common fabric (such as Frame Relay). Aggregate streams are TDM cross connected onto SONET. Trunk BW assigned based on peak rates. Hybrid Network Byte Aware Cross-Connect Fixed Rate Traffic: CSTDM bandwidth based on Peak Rates Bursty Traffic: Access lines aggregated onto higher load trunk. Packet Switch StatMux Trunks are CSTDM. Voice Quality vs. Bit Rate Quality G.729 G.728 G.726 G.711 G.723.1 8 16 32 Bit Rate (Kbps) 64 Switched Network Carrying Capacities High Speed Trunk Hybrid Carrying Capacity Cell Switch StatMux Packet Switch StatMux Circuit Switch TDM 0% Bursty 100% Fixed Rate Offered Traffic Mix 100% Bursty 0% Fixed Rate Switched Network Carrying Capacities Hybrid Network Carrying Capacity all bursty data traffic groomed onto packet network Hybrid Circuit Switch TDM 0% Bursty 100% Fixed Rate Offered Traffic Mix 100% Bursty 0% Fixed Rate Switched Network Carrying Capacities Hybrid Network Carrying Capacity Hybrid no data traffic groomed onto packet network 0% Bursty 100% Fixed Rate Offered Traffic Mix 100% Bursty 0% Fixed Rate Switched Network Carrying Capacities Hybrid Network Carrying Capacity real world network 0% Bursty 100% Fixed Rate Offered Traffic Mix 100% Bursty 0% Fixed Rate Switched Network Carrying Capacities Convergence Carrying Capacity Cell Switch StatMux Packet Switch StatMux Circuit Switch TDM 0% Bursty 100% Fixed Rate Offered Traffic Mix 100% Bursty 0% Fixed Rate 70’s & 80’s Fixed Rate Voice Dominates Data Voice 70’s & 80’s time Switched Network Carrying Capacities Convergence Carrying Capacity Circuit Switch TDM 0% Bursty 100% Fixed Rate Offered Traffic Mix 100% Bursty 0% Fixed Rate Turn of the Century A Mixed Traffic Environment Data Voice 2000 time Switched Network Carrying Capacities Convergence Carrying Capacity 0% Bursty 100% Fixed Rate Cell Switch StatMux Offered Traffic Mix 100% Bursty 0% Fixed Rate By 2005, Data Dominated Data Voice time 2005 Switched Network Carrying Capacities Convergence Carrying Capacity Packet Switch StatMux 0% Bursty 100% Fixed Rate Offered Traffic Mix 100% Bursty 0% Fixed Rate What's the impact of Video? Video #1 since 2010, is a packet switched statmux network best? Yes. Most video coders are variable rate. Two changes to make the network more video friendly… Might be a good idea to increase Ethernet's maximum packet size. All packets with bit errors shouldn't be dropped Voice/Video dropped packet = lower quality Better quality possible if payload delivered Carrying Capacity... Got bursty data traffic to move? A packet switched system using statistical multiplexing will allow you to service the most users given a fixed chunk of bandwidth. Got fixed rate traffic to move? A circuit switched system will allow you to service the most customers given a fixed chunk of bandwidth. WAN Trends 60's - Fixed Rate Voice Dominates Voice Network moving data on the side Mid to Late 90's – Mixed Traffic Environment New Carriers – ATM Older Carriers – Hybrid Early 00's - Mostly Bursty Traffic Data Networks moving voice on the side 10's - Mostly Video Data Networks moving video Data & voice on the side 1990 Marketing Glossy 1990 Marketing Glossy Example) Coding a Microphone Output m(t) volts (air pressure) time (sec) Energy from about 500 - 3,500 Hz. A/D Convertor m(t) volts (air pressure) 1/8000 second time (sec) Step #1) Sample the waveform at rate > 2*Max Frequency. Telephone voice is sampled at 8,000 samples/second. A/D Convertor Legacy Wired Telephone System uses PCM Pulse Code Modulation One of N possible equal length Code Words is assigned to each Voltage N Typically a Power of 2 Log2N bits per code word Wired Phone System: N = 256 & 8 bits/word Compact Disk: N = 65,536 & 16 bits/word A/D Convertor. 1 bit/sample. Example) N = 2. Assign 0 or 1 to voltage. 3.62 v, output a 1 t1 time (sec) 0 < Voltage < +5v, Assign Logic 1 -5v < Voltage < 0, Assign Logic 0 Bit Stream Out = 1111110000111... A/D Convertor. 1 bit/sample. Example) N = 2. Assign 0 or 1 to voltage. 0 < Voltage < +5v, Assign Logic 1 -5v < Voltage < 0, Assign Logic 0 Far side gets... 1111110000111 (13 samples) Needs to output 13 voltages. What does a 1 represent? A 0? Receive a 1? Output +2.5 v (mid-range) Receive a 0? Output -2.5 v (mid-range) Hold the voltage until next sample A/D Convertor. 1 bit/sample. Input to the transmitter. Output at the receiver. +2.5 v -2.5 v Considerable Round-Off error exists. A/D Convertor. 2 bits/sample Example) N = 4. Assign 00, 01, 10 or 11. 3.62 v, Assign 11 +2.5 v time (sec) t1 2.5 < Voltage < 5 , Assign 11 0 < Voltage < 2.5, Assign 10 -2.5 < Voltage < 0, Assign 00 -5 < Voltage < -2.5, Assign 01 -2.5 v Bit Stream Out = 11111011111100 000000101011... A/D Convertor. 2 bits/sample. Input to the transmitter. Output at the receiver. +3.75 v +1.25 v -1.25 v Receive 11? Output 3.75v Receive 10? Output 1.25v Receive 00? Output -1.25v Receive 01? Output -3.75v Reduced Round-Off error exists. -3.75 v Circuit Switched Voice (POTS) Bandwidth ≈ 3,500 Hertz A/D Converter samples voice 8,000 times/second rounds off voice to one of 256 voltage levels transmits 8 bits per sample to far side D/A Converter receives 8 bit code word outputs one of 256 voltage levels for 1/8000th second 64,000 bps (1 byte, 8000 times/second) Compact Disk Bandwidth ≈ 20,000 Hertz A/D Converter samples voice 44,100 times/second rounds off voice to one of 65,536 voltage levels transmits 16 bits per sample to far side D/A Converter receives 16 bit code word outputs one of 65,536 voltage levels for 1/44100th second 705,600 bps Sampling & Quantizing Examples fs = 16 KHz 4096 quantiles 256 quantiles (approximate phone quality) 32 quantiles 4 quantiles (generally 2 levels used!) 4096 fs quantiles = 16 KHz fs = 8 KHz (some interference) fs = 2 KHz fs = 1 KHz 1/8th Second of Voice 1/8th Second of Voice 1/8th Second of Voice Sources of POTS delay Source CO POTS TDM Trunk TSI Trunk resources are dedicated to each voice call via TDM. PCM Coder ... Local Loop PCM Coder POTS Local Loop TDM Trunk TSI Destination CO Intermediate Digital Voice Switches Sources of VoIP delay Voice Packet Transmission Coder Assembler Buffer Packet Switch ... Trunk resources are randomly assigned to each voice call via Statistical Multiplexing. Voice Decoder Receiver Buffer StatMux Trunks Intermediate Packet Switches Packet Switch Voice (Video) on LAN (WAN) More complex system than circuit switched voice Packet Assembler Transmitter Buffer Receiver Buffer End-to-End Delays > Circuit Switch TDM Delay Variability > Circuit Switch TDM