Implementation and simulation of Scheduling Algorithms in OPNET

1
Implementation and simulation of Scheduling
Algorithms in OPNET

• Project by Itamar Cohen
• Supervisor Nir Arad

2
AGENDA

• Introduction
• Applications
• Algorithms RR, WRR, WFQ
• Summary

3
MOTIVATION

• Scheduling algorithms are used in
• Computer networks, operating systems, real
  time applications.

4
MOTIVATION (cont.)

• Uses for nowadays networks

• Suppliers of services are committed to guarantee
  a fixed service level, which can be checked under
  a few pre-defined parameters.
• This is called SLA (Service Level Agreement).
• But the client would like to get much more
• A possibility to choose his favorite
• application, which will get a higher
  percentage of his bandwidth.
• This is called QoS (Quality of Service).

5
SCHEDULING ALGORITHMS

• Plenty of algorithms try to solve in different
  ways the problem of one server, which has to
  choose in real time the next client to be served,
  among a few clients.
• Each algorithm is good for some case, but bad for
  other cases. No algorithm is good for all the
  possible scenarios.
• The OPNET modeler gives us an excellent ability
  to test each algorithm under plenty of different
  scenarios.

6
THE PROJECTS AIMS

• This project implements OPNET standard packages
  for the following scheduling algorithms RR, WRR
  and WFQ.
• Each algorithm was simulated under a few
• interesting scenarios.
• The generic attitude of the implementation
  enables the user to simulate each algorithm
  under plenty of other scenarios.

7
Round Robin (RR)

• The simplest algorithm.
• The most important parameters which were checked
  are
• ETE Delay
• Backlog.

8
RR basic 2
9
RR - Interesting scenarios

• A source with smaller packets size
• A source whose packets are smaller then those
  of the rest of the sources will get a worse
  service The reason is that the RR is a
  packetized model, which doesnt consider
  different packets sizes.

10
RR different packets sizes
11
RR different packets sizes (cont.)

• The bad service causes increasing ETE Delay and
  backlog.
• Only after a long run the server takes advantage
  of the smaller packets, and stops this increase.

12
RR - different packets sizes ETE Delay
13
RR - different packets sizes - backlog
14
RR another problematic scenario

• Misses his turn
• When one source is unlucky enough to send its
  packets just after its turn in the current round
  passed, it will have to wait for a whole round
  till it will be served.
• When there are plenty of active sources, this
  wait time is not negligible at all.
• The ETE Delay of the unlucky source is a few
  times bigger then that of the other sources.
• The backlogs, however, are almost identical.

15
RR Misses his turn
16
Misses his turn ETE Delay
17
Misses his turn - backlogs
18
Weighted Round Robin (WRR)

• A more sophisticated algorithm, which solves the
  problem of one source faster.
• The user can promote different weights for
  different sources.

19
WRR
20
WRR different packets sizes

• If the user knows in advance the packets average
  size of each source, he can promote it to improve
  performances.
• The normalized weight is, therefore
• (int)(promoted weight / avg. packet size)
• Let us examine the effect on the scenario of one
  source with smaller packets size, demonstrated
  in the RR context.

21
WRR - different packets sizes (cont)
22
WRR - different packets sizes (cont)
23
WRR Pros.

• WRR solves the problem of one source, which sends
  smaller packets.
• WRR guarantees a higher bandwidth for the
  favorite sources, still without starving the
  other sources.

24
WRR Cons.

• The problem of misses his turn remains
  unsolved.
• And what will we do if the average packet size is
  not known in advance (or has a large STDEV)?

25
WFQ Weighted Fair Queuing

• GPS The ultimate choice!

• We would have preferred to handle a bit-by-bit
  (rather then packetized) weighted RR fashion.
• This is called Generalized Processor Sharing.
• Unfortunately, this is impractical.
• But we can approximate it.

26
From GPS to PGPS (WFQ)
27
Weighted Fair Queuing Overview

• WFQ schedules the packets according to their
  finish time had they been handled by a GPS
  algorithm.
• Parekh and Gallager had proved that WFQs
  performances are lower then that of GPS by only a
  small constant P G .

28
WFQ in different scenarios

• WFQ succeeds to give the same good results as WRR
  gave in the different packets sizes scenario,
  without requiring the user to promote the
  packets average sizes in advance.
• WFQ succeeds to solve the problem of Misses his
  turn, in which both RR and WRR failed.

29
WFQ not Misses his turn anymore !
30
WFQ Cons.

• The sophisticated scheduling requires higher
  computational complexity.
• Jon C.R. Bennett and Hui Zhang showed that under
  one specific scenario WFQs service is far AHEAD
  of GPS. This will result in unstable and less
  efficient network control algorithm B Z .
• To solve these problems, newer algorithm, such as
  W2FQ, were implemented

31
SUMMARY

• As the demands from computer networks become more
  and more specific and complicated, more
  sophisticated algorithms are invented
• Sophisticated algorithms are likely to give
  better performances, but they also require a
  higher computational complexity.
• There are plenty of newer interesting algorithms.

32
REFERENCES

• P G Abhay K. Parekh and Robert G. Gallager. A
  Generalized Processor Sharing Approach to
• Flow Control in Integrated Services Networks The
  Single-Node Case
• IEEWACM Transactions on nejworjong, vol. 1, NO.
  3, June 1993.
• Chuck Semeria. Supporting Differentiated Service
  Classes Queue Scheduling Disciplines.
• http//www.juniper.net/techcenter/techpapers/20002
  0.html
• Sridhar Iyer. Lectures slides for Autumn
  Semester, KR School of Information Technology,
• IIT Bombay. Queuing and Scheduling
  http//quark.it.iitb.ac.in/it605/lectures/sched
• Krishna Paul. Lectures slides for Autumn
  Semester, KR School of Information
• Technology, IIT Bombay. Scheduling
• http//quark.it.iitb.ac.in/it605/krishna/SCHEDULI
  NG.ppt
• Kevin Fall. Lectures slides for spring 1999, UC
  Berkley, EECS 122 - COMMUNICATION NETWORKS,
• Supplementary notes on WFQ
• http//www.cs.berkeley.edu/kfall/EE122/wfq-notes/
  wfq-notes.pdf
• B Z Jon C.R. Bennett, Hui Zhang. WF2Q
  Worst-case Fair Weighted Fair Queuing,