Abstract — Shishkin Y. S. — Cluster system for distributed modeling of computer networks

Shishkin Yevgeniy

master's site

DonNTU - DonNTU masters portal

Biography

Abstract

Library

Links

Search report

Individual task

Abstract — Shishkin Y. S. — Cluster system for distributed modeling of computer networks

Overview of existing researches

Today there are several software packets for computer modeling of computer networks. This is COMNET III by Caci Products Co., BONeS Designer by Cadence Inc., OPNET Modeler by Mil3 Inc, and NS2, based on VINT project.

All this systems are enough universally for simulation modeling of computer networks and allow defining exact results of modeling real networks for analysis and for receive practical results.

Deciding tasks

In my work I use simulation system NS2, because it is free and open source system. User can independently write any units and classes for work with simulation system. This system requires 250 Mb of HDD memory and 8 Mb of main memory. So, this system could be started even on Pentium 486. It's really easy, and you can organize parallel compute network with low-performance computers. Analogs of NS2 need's less of HDD memory, but more of main memory. You can use NS2 on almost all operation systems — Windows 95/98/Me/NT/2000/XP, Solaris, SunOS, Linux, FreeBSD, HP-UX.

I have a task — make analysis and compare two routing algorithms in wireless networks — AntHocNet and AODV. I need to make distributed modeling system, base on NS2, for distributed modeling of algorithms in wireless networks at cluster network, and also to model several standard networks and make analysis of results to define better algorithm for every specific network and for using this routing algorithms.

Theoretical analysis

In modeling of networks with NS2 system easy to use several computers in network. This is parallel distributed modeling. There is one feature — on every computer is computer networks started separate copy of NS2, which make modeling of some part of big network. There is an example of modeled network on picture 1.

Picture 1. Example of network, modeled in NS2. Animation, 8 frames, 58.5 Kb. Refresh page to replay animation.

Parallel distributed modeling using in cases, if you need to model large network and model it on large time interval, so you need large hardware resources. Parallel distributed modeling allows you save much time, as compared with modeling network on only one computer.

Plug-in PDNS — Parallel Distributed Network Simulator — used for realize of distributed modeling, coordination and synchronization of work. This plug-in use LibSynk library, there realized all functions for using of distributed modeling system.

Let's consider modeling routing algorithms in wireless networks — AntHocNet and AODV.

AntHocNet — it's ant-based algorithm, data packets equal to ants. The first step of this algorithm — starting of reactive forward ants, they initialize routing tables in every node. In addition, on every node computing value of "pheromone" — this is preference of using current link. Then, on second step, starting backward ants, they back to node, which initialize a connection. After this data transmission started, and also started proactive forward ants, they evaluate performance of transmission and search new paths. If node failed, starting repair ants to find an alternative path in place of failed link.

AODV (Ad-hoc On-demand Distance Vector) — algorithm, based on making routing tables on every node to minimize time of information transmission between nodes. At first this algorithm making routing tables on every node. In this table there is information about shortest way to every node thought every neighboring node. On first step in table there is information only about neighboring nodes. On every next step every node communicates with every neighboring node and gives information about shortest path to every node it known. After several steps routing tables become constant, then beginning data transmission through the shortest path. If node failed, taking the next shortest path from the table.

Personal developments

In development of this system was written set of packet executable files for start this system in Windows with Cygwin Unix-emulator. This packet files allow cooperating Windows console applications with console applications in emulator, and allowing run all processes for distributed modeling of computer networks.

Also has been written the program for union result files of different sub networks in one file, and analysis and visualize results of modeling different parts of network as results in united network.

I will write program shell for starting this modeling system on cluster network, using Windows Compute Cluster Server 2003. This shell will distribute input data thought the nodes, will start modeling process and will gather output results of modeling for further analysis.

Experimental researches

On realizing of this project for distributed modeling is planning to make experiments on cluster network to model wireless networks, using different routing algorithms.

There are plan to make 10-20 experiments for networks of different sizes — for small networks (on area 500х500 m) with small quantity of nodes (less then 15), for average networks (on area from 500х500 to 1500х1500 m) and average quantity of nodes (between 15 and 150), and for large networks (on area more 1500х1500 m) and large quantity of nodes (more then 150). I will make different series of experiments for every routing algorithm, so it will be possible to define advantages and disadvantages of every algorithm, and also define general appropriateness in wireless networks for further analysis and prognosis possible results on networks of another size.

Overview of results and summary

After making experiments I will define, which algorithm more preference for using in wireless networks of defining size, and also will define function appropriateness of wireless networks with using different routing algorithms for defining appropriateness and prognosis results for wireless network any size with any quantity of nodes.

This work is important for present community, which use wireless technologies, because it could gave him a real estimation of using tools and prognosis results of their work. Also this work could be used for defining more appropriate technical tools for every specific wireless network. It allows economize material and time costs in building real wireless network.

Perspectives of further researches

This research work isn't finished. On further development could be important improvement by upgrading components — packet files, distributed modeling system, cluster software, software shell etc.

Also using of this system for model any other routing algorithms in wireless networks or for modeling any wireless networks possible with any variety parameters, entered by system user. So, user should be used or realize necessary model and use it for further researches and analysis of this field of research.