Krzysztof Zatwarnicki

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.

Using adaptive fuzzy-neural control to minimize response time in cluster-based web systems

We have developed content-aware request distribution algorithm called FARD which is a client-and-server-aware, dynamic and adaptive distribution policy in cluster-based Web systems. It assigns each incoming request to the server with the least expected response time. To estimate the expected response times it uses the fuzzy estimation mechanism. The system is adaptive as it uses a neural network learning ability for its adaptation. Simulations based on traces from the 1998 World Cup show that when we consider the response time, FARD can be more effective than the state-of-the-art content-aware policy LARD.

Fuzzy-neural web switch supporting differentiated service

New designs of the Web switches must incorporate a client-and-server-aware adaptive dispatching algorithm to be able to optimize multiple static and dynamic services providing quality of service and service differentiation. This paper presents such an algorithm called FNRD (Fuzzy-Neural Request Distribution) which operates at layer-7 of the OSI protocol stack. This algorithm assigns each incoming request to the server with the least expected response time estimated using the fuzzy approach. FNRD has ability for learning and adaptation by means of a neural network feedback loop. We demonstrate through the simulations that our dispatching policy is more effective than state-of-the-art layer-7 reference dispatching policies CAP (Client-Aware Policy) and LARD (Locality Aware Request Distribution).

Neuro-fuzzy models in global HTTP request distribution

In the paper, the new method GARD of HTTP request distribution, in globally distributed clusters of Web Servers, is presented. The method described uses neuro-fuzzy models, enabling the estimation of the HTTP request response time. Both, the description of the method and the research environment as well as the obtained results are described in the paper.

Providing web service of established quality with the use of HTTP requests scheduling methods

The paper presents a new method for scheduling HTTP requests on the entrance of a Web service. An application of the described method enables keeping the quality of the Web services at a high, specified level, so as the response time for the whole WWW page would not exceed imposed time limit. In the paper, both the method and the research environment are described together with the results obtained in the simulation experiments.

Adaptive control of cluster-based Web systems using neuro-fuzzy models

Adaptive control of cluster-based Web systems using neuro-fuzzy models A significant development of Web technologies requires the application of more and more complex systems and algorithms for maintaining high quality of Web services. Presently, not only simple decision-making tools but also complex adaptation algorithms using artificial intelligence techniques are applied for controlling HTTP request traffic. The paper presents a new LFNRD (Local Fuzzy-Neural Adaptive Request Distribution) algorithm for request distribution in cluster-based Web systems using neuro-fuzzy models of Web servers in the decision-making process. The neuro-fuzzy model which is applied is discussed in detail and a design of the Web switch using the proposed solution is presented. Finally, a testbed is described and the results of a comparative simulation study on the LFNRD algorithm, and other algorithms known from the literature and used in the industry, are presented and discussed.

Global Distribution of HTTP Requests Using the Fuzzy-Neural Decision-Making Mechanism

This paper presents the HTTP request global distribution using the fuzzy-neural decision making mechanism. Two efficient algorithms GARDiB and GARDiB2 are proposed to support HTTP request routing to the Websites. The algorithms use the fuzzy-neural decision-making method to assign each incoming request to the Website with the least expected response time. The response time includes the transmission time over the network; as well as the time elapsed on the responding Website server. Simulation experiments showed that GARDiB performed slightly better than GARDiB2 and both proposed algorithms outperformed other competitive distribution algorithms in all simulated workload scenarios.

Global Adaptive Request Distribution with Broker

This paper presents the application of fuzzy logic and neural networks to HTTP request dispatching performed within a geographically distributed Web system. Web sites serve as a global content delivery system where each Web server can respond to the client request. We propose broker-based system architecture with a global request dispatching algorithm called GARDiB. The algorithm uses the fuzzy-neural decision-making mechanism to assign each incoming request to the server with the least expected response time. The response time include the transmission time over the network, both for the request and for the response, as well as the time elapsed on the server responding to the request. We demonstrate through the simulations that our algorithm is more effective than popular global dispatching policies Round-Robin and Weighted Round-Robin.

Operation of Cluster-Based Web System Guaranteeing Web Page Response Time

Guaranteeing Quality of Web Services (QoWS) is now very important for development of internet services. This paper describes a MLF (Mostly Loaded First) method of HTTP requests scheduling, and distribution in a cluster-based Web system. Its application enables keeping the quality of the Web service on the required level and makes the Web service behave in a predictable way, unlike to other Web services offering unpredictable and unreliable services of the best-effort type. The proposed system keeps the page response time within established boundaries, in such a way that at a heavy workload, the page response times for both small and complex pages, would not exceed the imposed time limit. We show thought simulation experiments the operation of the MLF system and the changes of distribution strategy while the load of the system increases.

Identification of the Web Server

The article considers the problem of modeling the Web server operations. At first, the simulation model of the server and the way of conducting experiments making possible to obtain required simulation model parameters of real Web server is introduced. The experiment resulsts and calculated values of parameter of specified Web server are presented. In the end the program written with use of the CSIM 19 package simulating operations of the Web server is discussed.