Leszek Borzemski

MWING: A Multiagent System for Web Site Measurements

When browsing the Web, users demand the low latency, high throughput and high availability of Web pages. Therefore each Web site should be measured, tested and evaluated from the users perspective. In the paper, we present the architecture of the system MWING that is able to test, measure and diagnose Web sites from the user point of view. It is a new generation system that adds new functions and extends the functionality and performance of the previously developed WING tool. MWING is now the multiagent measurement platform able to cooperate with independently developed measurement agents. Here we present how the MWING works and how the new architecture meets the ultimate user needs.

THE USE OF DATA MINING TO PREDICT WEB PERFORMANCE

Web mining is the area of data mining that deals with the extraction of interesting knowledge from World Wide Web data. The purpose of this article is to show how data mining may offer a promising strategy for discovering and building knowledge usable in the prediction of Web performance. We introduce a novel Web mining dimension—a Web performance mining that discovers the knowledge about Web performance issues using data mining. The analysis is aimed at the characterization of Web performance as seen by the end users. Our strategy involves discovering knowledge that characterizes Web performance perceived by end users and then making use of this knowledge to guide users in future Web surfing. For that, the predictive model using a two-phase mining procedure is constructed on the basis of the clustering and decision tree techniques. The usefulness of the method for the prediction the future Web performance has been confirmed in a real-world experiment, which showed the average correct prediction ratio of about 80%. The WING (Web pING) measurement infrastructure was used for active measurements and data gathering.

Distributed Web Systems Performance Forecasting Using Turning Bands Method

With the increasing development of distributed computer systems (DCSs) in networked industrial and manufacturing applications on the World Wide Web (WWW) platform, including service-oriented architecture and Web of Things QoS-aware systems, it has become important to predict the Web performance. In this paper, we present Web performance prediction in time and in space by making a forecast of a Web resource downloading using the Turning Bands (TB) geostatistical simulation method. Real-life data for the research were obtained in an active experiment conducted by our multi-agent measurement system MWING performing monitoring of a group of Web servers worldwide from agents localized in different geographical localizations in Poland. The results show good quality of Web performance prediction made by means of the TB method, especially in the case when European Web servers were monitored by an MWING agent localized in Gliwice, Poland.

A fuzzy adaptive request distribution algorithm for cluster-based Web systems

This paper presents a novel algorithm for distribution of user requests sent to a Web-server cluster driven by a Web switch. Our algorithm called FARD (fuzzy adaptive request distribution) is a client-and-server-aware, dynamic and adaptive dispatching policy. It assigns each incoming request to the server with the least expected response time, estimated for that individual request. To estimate the expected response times FARD uses the fuzzy estimation mechanism. With respect to the requirement of modifiability of the model, FARD uses a neural network provided with innate abilities for learning and adaptation. We implemented a prototype FARD-based Web switch that was used in experiments carried out to compare its performance to well known representative request distribution algorithms. The measurements show that FARD benefits can be significant, especially for heterogeneous Web clusters.

THE EXPERIMENTAL DESIGN FOR DATA MINING TO DISCOVER WEB PERFORMANCE ISSUES IN A WIDE AREA NETWORK

In this article we present the application of data mining (DM) to web performance forecasting. The main emphasis is to present the experimental design and data preparation steps of web performance forecasting based on web performance mining (WPM) method and developed in a wide area network using a multi-agent monitoring system MWING.

Data mining in evaluation of internet path performance

One of the difficulties of using Artificial Neural Networks (ANNs) to estimate atmospheric temperature is the large number of potential input variables available. In this study, four different feature extraction methods were used to reduce the input vector to train four networks to estimate temperature at different atmospheric levels. The four techniques used were: genetic algorithms (GA), coefficient of determination (CoD), mutual information (MI) and simple neural analysis (SNA). The results demonstrate that of the four methods used for this data set, mutual information and simple neural analysis can generate networks that have a smaller input parameter set, while still maintaining a high degree of accuracy.

BUSINESS-ORIENTED ADMISSION CONTROL AND REQUEST SCHEDULING FOR e-COMMERCE WEBSITES

The article concerns the problem of quality of Web service (QoWS) in e-commerce with the focus on request admission control and scheduling from the profit perspective of the owner of an e-business site. Two new functional extensions for a Web server system supporting business-to-consumer (B2C) activity are proposed. The first new feature is an ability to differentiate the key customers of the Web store and to offer higher QoWS to more valued ones. The second new function concerns offering high QoWS to all customers finalizing their purchase transactions. A new approach to B2C service differentiation using RFM (Recency, Frequency, Monetary Value) analysis is presented. A simulation model of the B2C Web server system and some results of the system performance for our approach in comparison with first in–first out (FIFO) scheduling are discussed.

Application of Transfer Regression to TCP Throughput Prediction

In this paper we present the application of the Transform Regression (TR) algorithm to predict file transfer time. We consider the files that are transferred from Web servers using HTTP protocol. We identify parameters having impact on file transfer time and use them as inputs to transform regression. We compare the results of TR-based prediction with the moving averages methods.

Architecture of Multiagent Internet Measurement System MWING Release 2

Internet performance measurements have motivated the development of a measurement system called MWING. Recently the existing system has evolved according to new requirements and now also distributed measurements are handled in Release 2. MWING system Release 2 is a multi-agent system that is used to perform tests, measurements and diagnosis of Internet and Web performance from the end user perspective. Multiple agents freely scattered in the Internet can, in synchronous or asynchronous manner, probe collectively given set of Web sites, and collect mutual performance data. This data can be used in the analysis of Internet performance e.g. for the needs of Content Delivery Networks and Grids. This kind of measurements is also desired and helpful for Internet Service Providers since it enables monitoring and analysis how ISPs domain is perceived by end-users from various Internet locations.

CLIENT-PERCEIVED WEB PERFORMANCE KNOWLEDGE DISCOVERY THROUGH TURNING BANDS METHOD

Client-perceived web performance prediction is a challenging for the knowledge engineering community. This article presents the application of the turning bands (TB) method to discover knowledge about the client-perceived spatiotemporal web performance. The TB method is used to study spatial or spatiotemporal phenomena. We conducted an experimental study to apply this technique. The MWING system was used to collect performance measurement for four clients, each monitoring the same group of web servers during several weeks of observations. In this article we present TB-based web performance prediction for an agent localized in Wrocław, Poland, and monitoring European web servers. The results of the spatial prediction of the total download web page times within a forecast of a two-week period showed that the TB simulation is efficient.