Outline Lecture 8 Performance Measurements and Metrics Performance Metrics Performance Measurements Kurose-Ross: 1.2-1.4 (Hassan-Jain: Chapter 3 Performance Measurement of TCP/IP Networks ) 2010-02-17 Sid 2 David Gundlegård, ITN Performance Metrics Performance Metrics Asessment of communication network performance How good is our network? (How much better than competitors?) Classical: Internet access bandwidth What can it be used for? Online gaming with HSDPA-connection? Vehicle safety communication via LTE networks? What shall we measure? How? Which metrics are important in different applications? 2010-02-17 Sid 4 David Gundlegård, ITN
Performance Metrics Performance Metrics Main quantitative metrics for packet switched networks Packet loss Due to congestion and bit errors End-to-end delay From sender to (final) receiver Delay-variation (A.k.a. Jitter) Variation in delay in a packet session Throughput (Bandwidth) Bits per second that can be sent through the network Other aspects Fairness CPU usage Memory usage Battery consumption Response time Availability Packet reordering Number of users Blocking rate Qualitative metrics Service experience: Good sound, bad picture, slow response 2010-02-17 Sid 5 David Gundlegård, ITN 2010-02-17 Sid 6 David Gundlegård, ITN Switching and Delay Switching and Delay Circuit switching Connection establishment delay Propagation delay Processing delay (typically small) Transmission delay Packet switching Store-and-forward delay Propagation delay Processing delay Transmission delay Queuing delay 2010-02-17 Sid 7 David Gundlegård, ITN 2010-02-17 Sid 8 David Gundlegård, ITN
Link Bandwidth and Delay Link Bandwidth and Delay Left: high bandwidth and/or propagation delay Right: low propagation delay and/or bandwidth Bandwidth x delay product Link utilisation and max throughput R max =W/RTT R = datarate W = window size Sliding Window flow control Utilisation Stop and wait 1 U = 1+ 2a W 1+ 2a Sliding window U = ( W < 2 a + 1) t a = t prop frame 2010-02-17 Sid 9 David Gundlegård, ITN 2010-02-17 Sid 10 David Gundlegård, ITN Delay Store-and-forwarddelay Prop. to packet size (L) In a switch: L/R Route with Q links: QL/R Cf. Cut through switch End-to-end delay (PS) Store-and-forward: d=ql/r Propagation: d=distance/velocity ~3*10^8 m/s (factor 0.7-1 depending on transmission media) Processing: time for process on every node proportional to the number of hops Error control, routing decisions etc. Transmission: d=l/r Time to transmit the bits on the link/path Determined by throughput (bottleneck) Compare with store-and-forward Queuing delay 2010-02-17 Sid 11 David Gundlegård, ITN 2010-02-17 Sid 12 David Gundlegård, ITN
A B Queuing Delay Queues occur when incoming traffic to a node (interface) is larger than the forwarding capacity (for some period of time) Forwarding capacity VS link/interface capacity Number of links/interfaces on a node 100 Mb/s Ethernet queue of packets waiting for output link 2010-02-17 Sid 13 David Gundlegård, ITN statistical multiplexing 1.5 Mb/s C Queuing Delay La Traffic intensity: ρ = R a = average rate that packets arrive to the queue L = packet length, R = datarate (service rate, capacity) What happens when La/R>1? Infinite buffer size? Finite buffer size? Analysed with Queuing theory Simulation 2010-02-17 Sid 14 David Gundlegård, ITN Queuing Models M/M/1 with and without priority The Markov assumption End-to-end Delay Metrics Round trip time (RTT) The time needed to travel from source to destination, plus the time to travel from destination back to the source. One way delay The time needed to travel from source to destination, or from destination to source Delay variation 2010-02-17 Sid 15 David Gundlegård, ITN 2010-02-17 Sid 16 David Gundlegård, ITN
Delay Variation (Jitter) The end-to-end delay variation over time (between packets) Important in real-time applications (buffer size and playout delay) Defined for a session of packets (>2) Max difference in delay Mean difference in delay Standard deviation... Throughput Number of bits per time unit that can be pushed through the link/network (A.k.a. bandwidth: ambiguity ) Data rate Often referred to as max transmitting rate at a link Throughput The actual data rate at the (final) receiver Taken into account flow control, bottlenecks, retransmissions, FEC, cross traffic (other users) etc. Throughput variation 2010-02-17 Sid 17 David Gundlegård, ITN 2010-02-17 Sid 18 David Gundlegård, ITN Packet Loss Percentage of packets lost Packet loss distribution Bursty Uniform TCP VS UDP Reasons Buffer overflow (congestion) Bit errors (packet discarded) Typical in wireless networks Performance Measurements 2010-02-17 Sid 19 David Gundlegård, ITN
Measurement Tasks Data collection Typically raw data from live network Analysis Statistical analysis of data Use collected data in simulation Presentation Visualisation through graphs and charts Interpretation What do the results tell us? We need new equipment Ip telephony can/cannot be used in current network Measurement Tools Monitoring tools Monitors existing traffic Ethereal, Tcpdump, Tcpstat Benchmarking tools Generates traffic used for analysis Often both client and server needed TPtest, Iperf, Netperf, Netpipe, DBS Standard tools Ping, traceroute, netstat 2010-02-17 Sid 21 David Gundlegård, ITN 2010-02-17 Sid 22 David Gundlegård, ITN Analysis Approaches Live network measurements Often difficult/expensive Does the system exist yet? Lab experiments Requires generalisations to more realistic conditions Simulations Time consuming/validation/verification Analytical evaluation E.g. queuing theory Quick but often many simplifications Combination For validation purposes Data Analysis Number of measurements Purpose of measurements? Variation in results? Length of every measurement Transient behaviour? Variability? Assumptions and configurations Generalisations possible? Statistical measures Average, min, max, standard deviation, variance, confidence intervals, hypothesis trial, correlation etc. 2010-02-17 Sid 23 David Gundlegård, ITN 2010-02-17 Sid 24 David Gundlegård, ITN
Internet Example Measurement Examples Ethereal >tracert IP/domännamn www.liu.se www.mit.edu >ping www.dn.se n 10 l 1000 TPtest5 Iperf 2010-02-17 Sid 25 David Gundlegård, ITN 2010-02-17 Sid 26 David Gundlegård, ITN Next Lecture Quality of Service 2010-02-17 Sid 27 David Gundlegård, ITN