Effective Packet Analyzing And Filtering System For Atm Network

EFFECTIVE PACKET ANALYZING AND FILTERING SYSTEM FOR ATM NETWORK

CHAPTER ONE

1.   Introduction

The context of the problem which this project addresses is how a network service provider such as US West or MCI can assign network bandwidth to a customer prospect, based on an understanding of typical network traffic originating with that customer. If the traffic under consideration was “telephone traffic”, the assignment of bandwidth is straightforward. In fact, many of the existing network formats were designed specifically with telephone traffic in mind. For example, a T1 ( or DS1) network connection can support a single telephone call on each of  its 24 channels. For various reasons, which will be described in the following sections,  “computer generated network traffic” represents a formidable challenge to both network service provider and the network customer. In particular, the “bursty” nature of computer generated traffic complicates the assignment of network bandwidth. Furthermore, the customers expectations on such things as “information loss” or “delay” of that information as it traverses  the network can be difficult to satisfy. This project focuses  specifically on  ATM  network  traffic, which is described in the next section. Important terms are defined later in thissection.

 

 

1.1.         Overview of Asynchronous TransferMode

Asynchronous Transfer Mode (ATM) could become the dominant means of transporting Wide Area Network (WAN) traffic within the next several years. It offers the flexibility to carry any type of traffic, whether constant bit rate (CBR), variable bit rate (VBR), or best efforts, i.e. available bit rate (ABR). ATM features cells having a fixed size of 53 bytes, including a 5 byte header and a 48 byte payload. Traffic is first chopped or "segmented" into cells at the sending end, and after being transported across the network, are "reassembled" at the receivingend.

 

Customer traffic, whether from a single customer, or the multiplexed aggregate from a number of customers, enters an ATM network via one of several input ports on an n x m ATM switch. In order to avoid swamping the network with traffic, it is reasonable to assume some sort of queuing structure as part of the ATM switch, router or similar equipment where traffic enters  the  network. ATM switches generally are constructed with identically sized queues on each output (and/or input) port. Conceptually, Figure 1 shows the traffic from multiple customers feeding  into a common ATMswitch.

 

 

Switch

Subnetwork

 

 

Figure1. End-to-end ATM network with multiplexed sources to a commonswitch.

 

1.2.         The Project Problem FurtherDefined

As an example, suppose we are given an 8-port ATM switch with traffic of 6 Mb/sec on each those ports.  One would expect that the ATM switch should have a capacity of 48 Mb/sec.   This is probably in the right ball park, but things are much more complicated due to the inherently bursty nature of network traffic. The primary function of an ATM switch is to combine or multiplex the traffic from all input ports to one or more output trunks. If traffic bursts from several input ports occur at the same instant, the switch may be overwhelmed with resultant loss of traffic (ATM cells). From the customer’s standpoint, such a loss would result in a  poor  Quality ofService.

 

Self-SimilarEthernetTraffic      Synthesized PoissonTraffic

Figure 2. Actual self-similar traffic vs. fictitious (synthesized) Poisson network traffic.

Characterizing the "burstiness" of ATM traffic is a particularly sticky challenge. Historically, network traffic had been assumed to obey a Poisson distribution. That is, traffic bursts are assumed to start at random times, and message traffic is assumed to be of arbitrary length. However, actual networkburstiness is worse than Poisson, and is characterized  variously as being self-similar, heavy-tailed, or long-term. This discovery was the result of recent (1993) research at Bellcore [Lela94], [Lau95]. Even the most casual observer can notice the marked difference between self-similar traffic and Poisson traffic, as shown in Figure 2, which is a reproduced here from the 1994 Bellcore paper.  A good description of the problem can be found in [Doul95].

1.3.     Important Definitions

 

1.3.1.  Burstiness

Burstiness relates to the non-uniform flow of network traffic. A channel which supports synchronous traffic at a constant bit rate is not bursty. One common metric of network traffic "burstiness" is the ratio of the peak-to-average bit rate for message traffic (as opposed to frame separators and idle frames). Since there seems to be no industry consensus for a single burstiness metric, several candidate metrics had to be evaluated as part of this project. For a discussion of burstiness see p.12 of Leland et. al. [Lela 94].

 

1.3.2.  Quality of Service(QoS)

For ATM networks, the two levels of performance to consider are: 1) the cell-level performance (cell loss and delay), and 2) call-level performance (call blocking as relates to Connection Admission Control (CAC)). No universally accepted metric exists for QoSon ATM networks, although work is being actively pursued in this area by the ATM Forum. In order to guarantee QoSto the user, it is necessary to introduce a set of QoSparameters whose properties indicate the nature and requirements in the layered protocol stack. For this project, the QoSfocus will be on cell-level performance, and key parameters are: 1) the ratio of cells lost to total cells offered for transmission, and 2) the delay distribution as cells traverse the network, end-to-end.

 

 

1.4          Objectives

The general objective of this project is the development of tools and algorithms which facilitate efficient use of ATM network resources and satisfy the customer’s requirements for quality of service (QoS). The software modules developed in this project should augment the library of  tools that can foster the research, education, and development efforts in the areas of network resource optimization and trafficmanagement.

 

A list of specific objectives includes:

• Extension of our existing literature survey for related work on ATM network optimization and trafficmanagement.
• Definition of network parameters suitable to the problem context for a typical elemental ATM network, which also must be designed.
• Definition of a set of metrics for evaluating the ATM network design. Of particular importance is the determination of a suitable metric for traffic “burstiness”, since there does not appear to be a commonly accepted metric in use today. Metrics relating to quality of service (QoS) should include cell loss ratio, bandwidth utilization, and end-to-end delay
• Development of a “toolbox” of script programs to process and manipulate network trace files suitable for input to the NIST ATM simulator, i.e. simulator pre-processing tools.
• Development of post-processing scripts to analyze simulation log files and extract QoS data and statistics. Script files for “pre” and “post” processing are a major deliverables of this work. The role which these scripts play in the simulation process is depicted in Figure3.

NISTATM Simulator

ATMROS Postprocessing QoS Script

• Development of an algorithm capable of predicting the ATM network bandwidth needed to carry traffic characterized by varying degrees of burstiness. The goal here is to be able to provide network administrators with efficient tools to plan or utilize more efficiently the available bandwidth in ATM networks and to provide reliable network services to theusers.

 

 

Figure3. Simulation flow, highlighting the role of ATMROS simulationscripts.

Sharing is caring!