p

Slide 1: 

ABSTRACT This project has two main goals: The first is implementation of a standard generic package of three scheduling algorithms in the environment of the OPNET modeler. The second is to simulate and analyze the results of these algorithms under several interesting scenarios, and compare the results to the theoretical results as appear in some articles from the last few years. The algorithms, which were implemented, are: Round Robin, Weighted Round Robin and Weighted Fair Queuing.

INTRODUCTION: : 

INTRODUCTION: The OPNET modeler: The model is based on the OPNET platform - a network simulation environment that enables simulation of different protocols, and different scenarios. The OPNET has many models based on different protocols built-in and included.

THE GENERAL MODEL: : 

THE GENERAL MODEL: The general model, which was used for all the different algorithms, which were implemented. Specific details related to each algorithm are given in this general model While implementing the models, the two main goals were: Create models, which are very generic. The scheduler of each algorithm can handle any reasonable number of clients, as far as the user promotes physical values, which keeps the system in the steady state (after initialization period). Maintain the models quiet simple and clear.

RR (Round Robin): : 

RR (Round Robin): This is the first algorithm which is used to describe about this opnet model The scheduler's process model: Specific data structures: the variable "next" points to the next client to be handled. According to the algorithm, "next" is initialized to "0", and is simply incremented by one (mod the number of the sources) on each transmitting event; if the q[next] is empty, "next" is incremented again. if all the queues are empty, the "lucky" queue, which sends the next packet becomes "next

Slide 5: 

simulations of different scenarios: In order to be able to analyze the results according to determined (non-random) behavior, all the parameters (packet size, packet interarrival time etc.) are constants and not statistic variables, unless specifically mentioned else. This is the simplest scenario, containing only two identical sources, which have the same packets sizes and the same rate.

SMALL PACKETS: : 

SMALL PACKETS: The Node Model contains 10 sources, from which 9 (sources 1 through 9) are identical. Source 0, however, generates packets, which are 10 times smaller, in a rate, which is 10 times faster (which means that its total bit generating rate is identical to that of the other sources). Here is the graph of the received packets:

MISSES HIS TURN: : 

MISSES HIS TURN: The scenario contains 10 sources having the same packet size and interarrival time. The only difference between them is the start time. Sources 1 through 9 are "lucky" enough to wake up each at the best time, which means – immediately after the server finished handling the previous source's (in the RR manner) packet. Source0, however, sends each packet in every turn a very little after its turn in the RR. Therefore he "misses his turn", and judged to wait for a whole RR turn till its packet will be served. The diagram for it is:

WRR (Weighted Round Robin): : 

WRR (Weighted Round Robin): The scheduler's process model : In aim to simplify the model, the weights promoted by the user are assumed to be already normalized to integer values. Note that this normalization contains division by the mean packet size of the source (which is assumed to be known in advance). The model accurately implements the models described by lyer [LYER] and in the site of Juniper [JUNIPER]. Special data structures – the "state" variable contains now also an array of integer – the sources' promoted weights.

Slide 9: 

simulations of different scenarios: small packets: This scenario is identical to (RR – small packets), and consequently the results are the same. This proves that generally using WRR with equal weights is identical to using RR (there might be some differences, however, because of the dependent in the (identical) integer weight which is given to each source). Misses his turn compensated: This scenario is similar to RR-misses his turn. The only difference is that here source 0 is given a weight, which is 10 times larger then that of any other source – and that, as a compensation for his “bad luck” to miss his turn.

Exponential with equal weights: : 

Exponential with equal weights: In this scenario there are 4 sources, which each of them sends packet in an exponential distribution with parameter 1 (per second). The packets’ sizes are all the same – 1024 bits. All the sources are given the same weight. Exponential with different weights: In this scenario sources 0 through 2 were given a weight of 1 each, while source 3 was given a weight of 10.

WFQ (Weighted Fair Queuing): : 

WFQ (Weighted Fair Queuing): The scheduler's process model: The model was implemented according to the lectures slides of fall [FALL], which accurately implement the weighted model of the Fair Queuing, originally proposed by Demers, Keshav and Shenkar [DKS]. Special data structures – note that here, in contrast to the case in WRR, the weights are doubles rather then integers. The new state variable, round_num, is also double. Also, we have a new field in the "intrpt" struct, which holds the time of the last interrupt; this is useful for calculating the round_num when the scheduler receives the following interrupt. We have also a new array, which holds the finish time of the last packet in every queue (this is used for calculating the finish time of the next packets which arrives to the queue).

Slide 12: 

simulations of different scenarios: small packets: This scenario is identical to 2.2.2 (RR – small packets). Specifically, note that in contrast to the WRR small packets’ scenario, source’s 0 smaller packets’ size isn’t assumed to be known in advance, and therefore source 0 isn’t given any “pre-determined privilege” on the other sources; the WFQ should be “smart enough” to compensate source 0 for its smaller packets size anyway! Misses his turn: This scenario is similar to RR-misses his turn.

Exponential with equal weights: : 

Exponential with equal weights: In this scenario there are 4 sources, which each of them sends packet in an exponential distribution with parameter 1 (per second). The packets’ sizes are all the same – 1024 bits. All the sources are given the same weight . Exponential with different weights: In this scenario sources 0 through 2 were given a weight of 0.1 each, while source 3 was given a weight of 1.0.

Slide 14: 

THANK U