Download Network Performance Course by David Tipper at University of Pittsburgh, Spring 2004 and more Study notes Communication in PDF only on Docsity!
TELCOM 2120Network Performance
David TipperAssociate ProfessorDepartment of Information Science and
Telecommunications University of Pittsburgh
Spring 04 TELCOM 2120: Network Performance (^2)
Course Objective
? To develop the modeling, mathematical andprogramming tools necessary to determine computer
systems and communication network performance.
? Performance Study: ÿ Estimating one or more metrics via
1. 2. measurement,simulation
ÿ Metrics are usually statistical quantities^ 3.^ analysis
− − Throughput, Response Time,Availability, Call Blocking, etc
Spring 04 TELCOM 2120: Network Performance (^3)
Course Outline? Course Organization, Introduction to Performance Modeling
? Review of Probability and Statistics ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗^ Reading : Jain 1, 2, 3 Reading : Notes, Jain 12
?? Performance Concepts, Collection/Analysis of MeasurementsAnalysis of Measurements, Reading: Notes, Jain 30, 12 Reading : Jain 13, 14 Comparison of Systems
?? LinearExperimental Design Regression, Experimental Design Reading : Jain 14, 16, 17, 20 Reading : Jain 20
? Simulation Concepts^ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ Reading : Notes, Jain 24
?? Basic Simulation ModelingSimulation Programming Reading : Handout, Jain 24 Reading : Notes
Spring 04 TELCOM 2120: Network Performance (^4)
Course Outline
? Simulation Input Modeling, Output Data Analysis Reading : Jain 25, 26, 27
?? Output Data Analysis, Validation of SimulationSimulation Software Reading : Notes, Jain 25
∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗^ Reading : Jain 25, 30
?? Basic Queueing ModelsQueueing Models and Networks Reading : Jain 30, 32, 33, Notes
? Queueing Networks, Applications^ Reading : Jain 32, 33, 35, Notes Reading : Jain 33, 34, 35, Notes
? Applications of Performance Techniques Reading : Notes
Spring 04 TELCOM 2120: Network Performance (^7)
History of Performance Modeling
? Orginated with Erlang in 1917 ÿ Studied circuit switched telephone networks – determine how
many operators to keep call blocking at reasonable level
? Kleinrock, Cerf studied data networks in 60’s ÿ Studied delay in ARPANET
? Sauer, Chandy – computer systems in 60’s ÿ Studied how disk drives interact with computer processing
power, time sharing strategies
Spring 04 TELCOM 2120: Network Performance (^8)
Metrics, Parameters, Factors
? Metrics ÿ Criteria to used to evaluate performance of a system
? Parameters ÿ System parameters
ÿ Workload parameters^ −^ h/w and s/w parameters.
? Factors^ −^ characteristics of user requests
ÿ ÿ Parameters to be varied during a study.Values can be varied to different levels (i.e., numerical values)
? Metrics are usually statistical quantities
Spring 04 TELCOM 2120: Network Performance (^9)
Performance Metrics
? Moments ÿ Response time
ÿ ÿ ThroughputUtilization
? Probabilities^ ÿ^ Queue length
ÿ ÿ Call blocking ratePacket loss rate
ÿ ÿ Packet out of order rateDistribution of response time
ÿ ÿ Bit error ratesPacket error rates
? Display as graph or table
Spring 04 TELCOM 2120: Network Performance (^10)
Performance Techniques
? Three basic techniques for performance analysis 1. Measurements
2. Simulation^ ÿ^ Gather Experimental Data from prototype or existing system
3. Queueing Theory^ ÿ^ Experiment with^ computer model of system
ÿ Analytical model of system
? Complementary tools ÿ Should cross check them
? Consider an example of each
Spring 04 TELCOM 2120: Network Performance (^13)
Measurements: Example
Spring 04 TELCOM 2120: Network Performance (^14)
Measurements: Example
Spring 04 TELCOM 2120: Network Performance (^15)
Measurements: Example
Tester
Spring 04 TELCOM 2120: Network Performance (^16)
Traffic Measurement Experiments
? Attach the broadband test system to the vBNS node at PSC
? Generate virtual circuit traffic with known characteristics, which islooped back to the tester, and measure QoS parameters (delay,delay-jitter, cell loss rate, etc.)
ÿ Repeat each experiment by varying − − the background traffic loadings.the number of hops that the tagged virtual channel is routed
− − through.the maximum rate of traffic for the tagged virtual channel.the traffic distribution of the tagged virtual channel.
? Capture application traffic and characterize it.
Spring 04 TELCOM 2120: Network Performance (^19)
Measurements: Example
Spring 04 TELCOM 2120: Network Performance (^20)
Simulation
?? Conduct experiments on a computer model of the systemStrengths
ÿ ÿ Allows details to be included.Can compare alternate system designs.
ÿ ÿ Can control time scale.Existing system not required.
? Weaknesses ÿ Difficult to generalize results.
ÿ ÿ Difficult to consider all cases/parameter values.Hard to determine sensitivity.
ÿ ÿ Time to develop and execute simulation.Effort to validate model and analyze output data.
Spring 04 TELCOM 2120: Network Performance (^21)
Simulation : Example
? Route Selection (Fault Tolerance Routing): ÿ Study effect of routing schemes on congestion
ÿ Analyze behavior of four routing algorithms^ −^ restrict area of restoration vs. spreading out
1. 2. Minimum DelayMinimum Hop Scheme
3. Load Distribution Among Paths^ ÿ^ Restrict path selection to minimize number of secondary nodes.
4. Load Distribution Among Links^ ÿ^ Choose path with maximum residual capacity.
ÿ Choose paths to equally balance load on outgoing links of primarynodes.
? Define appropriate cost formula for each scheme
Spring 04 TELCOM 2120: Network Performance (^22)
Simulation : Example
Model 10-Node Network
Spring 04 TELCOM 2120: Network Performance (^25)
Performance Evaluation
? Analyze N x N non-blocking output buffered switch
? Assumptions ÿ Arrival streams are independent
ÿ ÿ Bernoulli arrival processService time deterministic – D
ÿ ÿ Buffer size fixed – SSUniform distribution of traffic
Spring 04 TELCOM 2120: Network Performance (^26)
Performance Evaluation
? Define embeded Markov Chain at slot times πi,j = Prob{ i class ‘1’ cells, j class ‘2’ cells}
π πn = [= [ ππ 0 0,n , π 1 , π , π1,n-1 2 , … , , … , πk π ] n,0 ]
? Solve for steady-state probabilities π = π · P
where P is state transition matrix
Also use normalization condition π · e = 1 where eT = [ 1, 1, 1, …, 1 ]
? Exact form of P depends on space priority scheme modeled
KK − KK
P
α , α ,
α αα ααα αα α α α α 1
1 0 2 11 1 3 22 21 2 3 32 3 3 4 0 0 0 1 0 2 0 0 0 0 0
MM MM MM MM MM LO LO MM MM
MM M LLLLLL
LL
Spring 04 TELCOM 2120: Network Performance (^27)
Performance Evaluation
? No Priority Scheme ÿ Cells accepted into the buffer in FCFS fashion.
ÿ When buffer is full, all packets are rejected.
Spring 04 TELCOM 2120: Network Performance (^28)
Performance Evaluation
? Partial Buffering Scheme – (Nested Thresholds) ÿ Define a threshold Ti for each class i
ÿ ÿ If number in the systemHere two class [ T 1 , T 2 ] Set ≥ T iT ,all new class 1 = K i packets dropped
'
Spring 04 TELCOM 2120: Network Performance (^31)
Performance Evaluation
? Experiment 2 ÿ Define grade of service requirements
ÿ ÿ ∆∆ 12 Acceptable Loss Probability for class ‘1’ cellsAcceptable Loss Probability for class ‘2’ cells
? For specific traffic mixture (% class 1, % class 2) ÿ Determined maximum offered load (MOL)
Spring 04 TELCOM 2120: Network Performance (^32)
Performance Evaluation
Spring 04 TELCOM 2120: Network Performance (^33)
Choosing Performance Technique
?? Each method has range of applicabilityShould cross check results
? Table from Jain textbook
**7. Saleability6. Cost5. Trade-off Evaluation LowSmallEasy MediumMediumModerate HighHighDifficult
- Accuracy *3. Tools LowAnalysts ModerateComputer Languages DifficultInstrumentation
- Time Required1. Stage SmallAny MediumAny VariesPost-Prototype**
Criterion AnalyticalModeling Simulation Measurement
* In all cases, result may be misleading or wrong.
Spring 04 TELCOM 2120: Network Performance (^34)
Performance Study – Steps
1) Define goals and system
2) List system services and possible outcomes
3) Select performance metrics
4) List system parameters and workload parameters
5) Select factors to study and level values
