[Russian] [English]

Main

Abstract

Library

List of references

Results of searching in INTERNET

Individual task is translating site materials into English

Tereschenko Valeriy Vitalievich

Master of the faculty of Computer Information Technologies and Automatics of Donetsk National Technical University

Profession: Information Control Systems and Technologies

Subject of Master Work: Development of models and software products for the analysis and determining parameters of the work of Internet network nodes

Director of Master Work: Candidate of Technical Science, Reader Sergey Vladimirinich Lazdyn

E-mail: tereschenkoVV@rambler.ru

Abstract

Introduction

The subject of my master's degree work is “Development of models and software products for the analysis and determining parameters of the work of Internet network nodes”.

Internet is the net which connects millions of computers in the world. These computers may be of different types, different operational systems and other parameters. It has become possible thanks to TCP/IP (Transmission Control Protocol / Internet Protocol). Internet is the union of local networks, which are called Internet network nodes. They are joined among themselves. And they provide access for end-consumers. Connections can be different, starting with modem connections (lower than 50 bps) up to satellite communications systems.

The term “Internet network node” may have different definition. The first one is defined as the set of html documents, which are linked by hyperlinks. Secondly, it's a computer or a local network. Files and documents are situated there and it provides access to end-consumers. And at last, it's a software product, which is meant for access to documents' set using local or global network. Later on we will use the second definition of the term.

Internet network nodes consist of set of communication channels, which join it with other nodes and end-consumers, and own local network. We can see it at picture 1. Internet network nodes has to sustain heavy loading. Inasmuch as number of queries to the typical node rises steeply every month in 2-10 times, sometimes in 100 times for a 3-6 month. A node can start its non-stable work. We have the necessity of modelling and forecasting its work.

Picture 1. Internet network node.

(One click on the picture to animate and double click to stop animation)

1. Estimation of Internet network node parameters.

1.1. Estimation of Internet network node global parameters.

Response time of a website is used as a criterion of its performance. If company website is slow or unstable, it will negatively influence company image and sales volume. According to a research conducted by Zona Research (USA), 30% of visitors leave a web page if it loads for more than 8 seconds, and 70% of visitors leave disappointed if this time exceeds 12 seconds.

Beside integral test of the resource reaction they use many tests and parameters, which define the working of the node as a whole and work of the different constituents. We can divide it into the following groups: overall performance, Internet communication channel's productiveness, server's productiveness and network productiveness.

Websites' productiveness testing is intended for Internet network node optimization.

Websites' productiveness testing allows me the following:

-  to get objective appraisal of interdependence quality of the work depending on the clients' number, time, day of a week and others;

-  to value node's safety capacity depending on the number of clients;

-  to value website effectiveness design, as far as productivity is concerned;

-  to choose a good provider;

-  to optimize Internet network node behaviour;

-  to organize the monitoring followed Internet network node work quality.

Monitoring is performed with probes that are placed in different Internet locations (at different ISPs). Probe is a computer, when installed testing software programs. This software does the automatic looping executions to the Internet network node, emulate the user's actions and measure the server reaction time. Moreover, a server is periodically interrogated by ICMP-queries (“pings”) with predetermined rate. Probes can be placed on the different ISPs and in the user's local network. At least one probe (local probe) it is advisable to place in the ISP's local network of fault-testable server. Other probes (remote probes) are placed in the most interesting Internet points, which are considered to be the most interesting from access time view.

So as to revealing the factors, which exert most influence on the Internet network node work quality, the results of the tests are analysed by means of the special software programs. Probabilistic, correlation and regression analysis of data are use.

1.2. Determination and estimation of productivity parameters the parts of the Internet network node.

1.2.1. Determination of quality communication channels.

Determination of global Internet network node parameters sometimes haven't answers about reason of non-stable work. One of such reasons may scarcity or non-stable work communication channels.

Nowaday many Ukrainian ISPs, when they make contracts, consider technology (DSL, Frame Relay/ATM, RadioEthernet, satellite communications system) and bit rate(64 Kbps, 128 Kbps and so), but don't consider quality of them. But different network technologies have different consumes. The equidimensional communication channels may differ from quality: accessibility, response duration and so.

Most of foreign ISPs, when they are enlisted into contracts with corporation clients, i.e. covenant about Service Level Agreement (SLA) decide payment depends on the quality of the channel, it's the prevailing ideas of if.

Nowaday the most popular quality parameter of the communication channel is a “ping time” (time from sending ICMP-query till getting ICMP-answer) alongsize. With many advantages, it has a defect. Equipment processor responds to ICMP-query. But data are transferred in most cases through hardware.

So that if we want to state the value of communication channel, we have to define three parameters:

•  A ccessibility.

•  Max effective productivity .

•  Average response duration.

Nowaday we have two methods for the quality rating of communication channel. The first is the response duration at HTTP level. The second is the response duration at TCP level.

1.2.1. Servers' productivity.

Servers' productivity has an impact on Internet network node.

The main productivity parameters as for Internet network node are:

MS Windows/NT4/2000/XP servers

•  M ain memory size.

•  Average number of page in a second.

•  Percent of the paging file using.

•  Percent of the processor using.

•  Percent of the data queries.

•  Percent of the time needed for reading/writing.

•  Server downtime.

Database servers

•  Percent of the queries to the cash-memory.

•  Average number of page in a second.

•  Average number of pack-queries in a second.

•  Average number of recompiling in a second.

•  Server processor u tilization .

•  Agent processor u tilization .

•  Average waiting period

•  Average blocking queries waiting period.

MS DNS servers

•  Memory size as for nodes'.

•  Dynamic renewals number .

•  C all-back frequency .

•  Sending reports frequency .

FTP servers

•  Reading / writing speed .

•  A ccessibility.

•  Users' number

•  Average percent of the processor loading.

•  Max percent of the processor loading.

Email servers

•  Size of the email messages on the server.

•  Number of the email messages on the server

•  Number of the mail boxes.

•  Query frequency to the server.

•  Average volume of received mail size.

•  Average volume of sending mail size.

•  Max mail size.

•  Average percent of the loading.

•  Max percent of the loading.

1.2.3. Local network's productivity .

Every Internet network node has a local network, which involves servers, communication cables, modems, switches, hubs, routers, and other network equipment. We may use many software products, methods and technologies for the determination of working parameters of the network as a whole and its components. Internet network node local network parameters and possible faults can influence its work, but haven't any attitude to my master's degree work subject. I don't handle these problems in this work.

2. Modeling similar systems .

2.1. Total analysis.

Local network is more close system to the one under consideration. Later we will quickly view the main local network modelling methods. Nowaday for its using modern project methodology, analytical modeling and simulation technique. Modelling is one of the most popular analysis. Local network model is viewed as such which consists of some parts of information, which is created to study there. It usually uses abstract models, which represent the description of local network in some special language. Abstract model, which is represented by mathematical notation, means simulator.

2.2. A nalytical modeling

Using analytical methods is connected with necessity to construct local network mathematical models using the rigorous mathematical expressions. Analytical local network models usually have a stochastic nature and are built on the basis of queueing theory, probability theory and Markoff process and diffuse approximation. Differential and algebraic equations may be using .

2.2. Simulation technique.

In contrast to analytical modeling, simulation technique has more possibilities in image in modeling the process of functioning, dynamic interaction of components, complex intercommunication of fficiency parameters and factors of the system. Although simulation models are usually more labour-intensive, less laconical than analytical model they can be arbitrarily alike to modeling system and easy to use.

2.4. Experimental approach.

The prior methods are usually possible in use if we have information about system characteristics. This information can be go by "rule of thumb" methods. Nowaday on basis of these methods tools for investigation of local network hardware and software components are developing. Necessary information is go with the help of custom resources, which measure local network parameters. It's network analyzers and others.

2.5. Analytical modeling on the basis of queueing theory.

When we use analytical modeling, we replace investigation process or objects by investigation models. Identity of equations' forms and one-meaningness of interrelations between variables in equations, describing an origin and model are considered to be the basis of the method.

Conclusion

On the basis of gathered and above-mentioned material, we have come to the conclusion that a problem of the Internet network node modeling depending on communication channels number and types, different queries type with help on queueing theory and performance prediction wasn't handled. It's issue of the day. It was decided to change this line of action in the future investigation.

Queries come from the low speed channels (phone lines) and high speed channels (radiomodems, fibre-optic cables, satellite communication systems and other). The most difficult in this model is a heterogeneity communication channels and queries. Modeling queries in-service time need using different distribution laws to the different server-program's types.