Discrete Event Simulation: Tools and Applications

1
Discrete Event Simulation Tools and Applications

• Pawan K. Choudhary
• Kishor S. Trivedi
• Center for Advanced Computing and Communication
• Department of Electrical and Computer Engineering
• Duke University

2
System Analysis

• Two Approaches
• Measurement of system- Expensive, time consuming,
  but most accurate
• Modeling of system
• Analytical Modeling - Fast, Some cases not
  possible due to complexity in computation.
• Closed form Solution- Exact, no errors
• Numerical Solution- Numerical errors, round off
  etc.
• Simulative Modeling- Less Assumptions, models
  complex system, better insight of system before
  actual implementation.

3
Discrete Event Simulation

• In this state variable changes at discrete points
  in time, e.g., length of a queue in a packet
  switching network or buffer occupancy
• Most of present simulation tools are discrete
  event simulators which can be broadly classified
  into
• General Purpose Programming Languages C, C,
  Java
• Plain Simulation Languages- Simula, Siman, GPSS,
  etc.
• Simulation Packages- OPNET, CSIM-19, ns-2,
  GloMoSim, GTNetS etc.
• In following slides, different computer and
  communication applications analyzed in
  OPNET,CSIM-19 and ns-2 are presented.

4
CSIM Example- Performance of Client-Server in
Token Ring arrangement

• A Process-oriented Discrete Event Simulator.
• Implemented in C/C as a library of routines
  which implement necessary operations.
• Models a system as a collection of CSIM processes
  that interact with each other by using the CSIM
  structures.
• CSIM is patterned after the simulation language
  ASPOL
• More information and documentation can be found
  at http//www.mesquite.com/

5
CSIM Example- Performance of Client-Server in
Token Ring arrangement

• Problem statement
• N workstations and one file server are arranged
  in ring topology and token moves from station to
  station interconnected by LAN.
• Each client transmits by grabbing the token,
  while others just waits for their turn.
• Also client cannot transmit a message unless it
  has received all pending replies.
• Similarly server replies to client only when it
  grabs the token.
• Message generation, service time distribution,
  transmission delays, all are derived from
  exponential distribution.

6
CSIM Example- Performance of Client-Server in
Token Ring arrangement

• Parameters
• Client generates a request at rate ?
• Time to transmit request is 1/µ 0.1
• Time to transmit reply message ?1/? 3.2 as
  reply packet length has mean of 4KB
• Time required to process a request by server ?1/?
  2.0
• N5 clients are taken
• 1/? (0.01/(N1) 0.0024)
• ?? (N(µ ?)/?) where ? is offered load.
• The response time in milliseconds is plotted
  against the offered load

Comparison with Analytical model 1 shows
results match closely and difference for all
values is less than 10 all the time 1Oliver
C. Ibe, Hoon Choi, Kishor S. Trivedi, Performance
evaluation of Client-Server Systems, IEEE
Transactions on Parallel and distributed
systems, Vol.4 , No.11, November 1993.
7
OPNET Modeler example- Comparison of RED vs. FIFO
with Tail-drop

• Object Oriented Simulation tool with hierarchical
  model structure.
• Menu Driven packages with Graphical User
  Interface.
• Contain TCP/IP protocols and applications, MPLS,
  IP QoS, RSVP.
• Wide range of network equipment from different
  vendors like CISCO, 3Com, HP etc.
• Can run on different operating systems
• Can support concurrent users
• Has built-in graphing tool for output analysis,
  animation capabilities.
• More information can be found at
  http//www.opnet.com

8
OPNET Modeler example- Comparison of RED vs. FIFO
with Tail-drop

• The network for the example consists of two
  routers and three clients with their
  corresponding servers.
• The capacity of link between two routers is taken
  to be 2.048Mbps.
• All other links have capacity of 100Mbps fast
  Ethernet.
• Clearly the link between Router A and Router B is
  the bottleneck.
• Our goal is the queue length at Router A for the
  two schemes.

9
OPNET Modeler example- Comparison of RED vs. FIFO
with Tail-drop

• The application chosen is video conferencing with
  incoming and outgoing frame sizes are set to 1500
  bytes.
• Notice that both queues using RED and FIFO
  taildrop behave similarly when link utilization
  is low.
• After 40 seconds, when utilization jumps to
  almost 100 , congestion starts to build at
  router buffer that uses FIFO taildrop.
• In case of active queue management (RED case),
  the queue length remains low and it never
  saturates.
• In fact queue length is much smaller than that of
  FIFO during the congestion period

10
ns-2 Network Simulator

• Started as variant of REAL network simulator with
  the support of DARPA and several
  companies/universities.
• Public domain simulation package in contrast to
  OPNET MODELER/CSIM which are commercial packages.
• Like OPNET MODELER/CSIM, it also uses an object
  oriented approach towards problem solving.
• It is written in C and object oriented TCL.
• Unlike CSIM, ns-2 is (network) domain oriented.
• Details about ns-2 can be found from
  http//www.isi.edu/nsnam/ns/

11
Web Cache example in ns-2

• We are interested in file delivering latency or
  response time.
• It is defined as time interval from the browser
  clicking on the object to the requested object
  being displayed on the monitor
• When some hyperlink is clicked, several URL
  requests are sent from browser to Web proxy,
  which has cache of URL.
• If consistent copy of file is present then an hit
  occurs otherwise miss occurs.

Typical infrastructure interconnecting a subnet
with internet.
12
Web Cache example in ns-2

• Simulation is carried out for 100 browsers
  logging into the internet simultaneously.
• The interarrival time is derived from Poisson
  process and service time is Generally distributed
• The document size is derived from Pareto
  distribution.
• Three cases are compared
• Bandwidth 256kb/s, with no web cache.
• Bandwidth512 kb/s, with no web cache.
• Bandwidth 256kb/s, with web cache having hit
  ratio 0.5.
• We see that the configuration with the web cache
  and half the bandwidth (256kb/s) performed almost
  as well as, and in some cases better then, the
  configuration with greater bandwidth (512 kb/s)
  and no cache.

1.Series 1 is 256kb/s with no cache 2.Series 2 is
256kb/s with 50 hit ratio 3.Series 3 is 512kb/s
with no cache

• Cache is generally much less expensive
• and easier to add to a network than
• bandwidth, hence using cache is more
• effective solution.