An Optimization Technique of Web Caching using Fuzzy Inference System

International Journal of Computer Applications (0975 – 8887)
Volume 43– No.17, April 2012

An Optimization Technique of Web Caching using
Fuzzy Inference System

Anish Kumar Saha Partha Pratim Deb Moutushi Kar, D. Rudrapal
Assistant Professor M.Tech CSE Assistant Professor
CSE Department Netaji Subhash Engg College CSE Department
NIT, Agartala, INDIA West Bengal, India NIT, Agartala, INDIA

ABSTRACT Three main important issues that have profound impact on
Caching and Prefetching are the two approaches for Web caching management namely: Cache Replacement, Cache
Caching. Again Web caching is a technology to reduce the Consistency and Cache Prefetching [2].
response time, bandwidth uses and improving the network Cache replacement algorithm is required when cache is full and
traffic etc. Web Prefetching tries to put the future used web need to delete or replace some web objects to give place for the
objects into cache with higher probability of cache hit. In Web new coming web objects. The cache replacement is the core of
caching, Cache replacement algorithm is the core of it. So, the web caching; hence, it is very necessary to create a good
good replacement policy would make effective management of replacement algorithm for caching mechanisms achievement.
cache memory utilization with higher probability of cache hits. In general, cache replacement algorithms are also called web
General replacement policy like LRU, FIFO, LFU considering caching algorithms [2]. There are some of the important cache
only the arrival time, but other parameters related to web replacement policies: first-in/first-out (FIFO), least recently
objects should consider for deciding cacheable or not. This used (LRU), least frequently used (LFU), and so on. They
paper approaches a replacement policy with fuzzy inference consider only one factor, which is Recency or arrival time of
system with input parameters Frequency, Latency and Bytesent the web objects.
of web objects. By considering these parameters, the Other than Recency parameter, other parameters can be the
replacement would have artificial intelligence in cache page size, fetching delay, reference rate, invalidation cost,
replacement policy. invalidation frequency of a Web object etc [15]. These
parameters should consider whenever a good replacement
algorithm required. Using these parameters into their designs,
Keywords these cache replacement algorithms shows good performance
Web Caching; FIS (Fuzzy Inference System); Frequency; improvement over the conventional ones like LRU, FIFO,
Latency; Bytesent. LFU, and so on.
Then come to the Cache consistency. It is concern to determine
1. INTRODUCTION AND RELATED that the contents in the cache should be up to date when the
WORKS objects get changed on origin server. That means the original
sever contents and proxy contents for some web objects should
Users of internet are increasing day by day. But the bandwidth
be same. Most frequent changing web objects are not good for
is limited. As more space of our bandwidth is used, latency of
Web Caching due to its frequent inconsistency state. Cache
documents retrieved from the Internet increases. Despite the
consistency is classified into 2 groups. One is weak cache
fact that the web is growing fast, the same documents get
consistency (Less freshness) and another strong cache
requested and the same web sites are visited repeatedly. We
consistency (High Freshness). Different examples of Weak
can take advantage of this to avoid the downloading redundant
consistency are TTL (Time-To-Live), PCV (Piggyback Client
objects. That’s why web caching is introduced to place the
Validation), PSI (Piggyback Server Invalidation) etc. Different
redundant objects closer to the clients that are likely to be used
examples of strong consistency are Polling-every-time,
in the near future. But, we need to use the cache memory of
Invalidation information from server, Lease in Client/server
proxy server [12] in a convenient way to get the good result of
Interaction [3]. So, Which Cache Consistency algorithm would
web caching.
be applied is fully based on the application type and its needs.
Web pages can be cache at Client, Proxy Server and Original
Now comes to the point, Prefetching, which aims at predicting
Server. Based on the location, caching can be Browser
future requests for Web objects and bringing those objects into
Caching, Client Side Proxy Caching, or Server Side Proxy
the cache in the background, before an explicit request is made
Caching [1]. Browser caching is economical and effective way
for them. Web caching & Web Prefetching can complement
to improve the performance of the World Wide Web due to the
each other since the Web caching technique exploits the
nature of Browser Cache, which is closer to the user. In Client
temporal locality, whereas Web Prefetching technique utilizes
Side Caching, the web objects are placed near to the clients.
the spatial locality [1] of Web objects. Different types of
And in server-side web caching, the cache pages is near to the
Prefetching methods are there. These are Prefetching
server to reduce the Web server’s load and, thus, to shorten the
hyperlinks, Association relation, History based, Page rank
user perceived response time.
based [1].

20


Except these, Web log mining is also used for Prefetching [5]
[6] [7] [8] and other adaptive Prefetching algorithms are Where are input variables,
already there [2]. So, we can conclude that web caching and
are fuzzy sets and y is either constant or linear
Prefetching both are the two techniques use to reduce the
response time perceived by the user. Recently artificial function of input variables.
intelligence is applied in almost every field to get human like In this proposed model, parameters or inputs used here in the
thinking work done by machine. Fuzzy logic, Neural Network, FIS are Frequency means number of cache hits for particular
Genetic Algorithms all is having their roots in Artificial period of time, Latency is the fetching delay or retrieval delay
Intelligence. from server and Bytesent is the size of the web object or byte
sent from original server to proxy server. Fig 1 Show the FIS
structure.
2.PROPOSED ARCHITECTURE
Only one parameter Recency or arrival time is not sufficient for
good replacement of web objects from Proxy Server Cache
memory. Some examples of arrival based cache replacement
algorithm are LRU, LFU, and FIFO etc. Here three parameters
of the web objects are used for cache replacement policy.
These are Frequency, Latency and Bytesent. The First
parameter is Frequency. Frequency mean number of cache hits
of the web objects for a period of time in the proxy server.
Higher Frequency value would give higher probability of cache
hits and providing quick retrieval of web objects to clients from
Proxy Server. The second parameter is Latency. Latency means
time to get the requested web object from server. If the web
objects required higher Latency, it would be given higher
priority of cache. Because, if the high Latency web objects are
place into cache of proxy server, the repetitive client requests
would be deliver quickly to the clients from proxy server thus
providing quick retrieval. The third parameter Bytesent means
the size of the web object or the byte sent by the server to the
Proxy Server for the requested web object. The Bytesent or the
size of the object is crucial for replacement policy. If the object
is more size, then it would be given high priority of cache,
because higher web object size required more bandwidth for Fig 1: FIS and its input parameters and output
transmission. So, it would be better to place higher size web The membership function of each input is divided into two
objects into Cache of proxy server which is near to the clients regions namely, small, and large. Here all the membership
and serving repetitive client request from proxy server, thus functions are bell shape member function. The curves of
reduces bandwidth. membership functions are shown at fig 2.
The proposed architecture cab be explained as
 When the users navigate web pages, all web objects are
placed in Proxy Server cache primarily. The information
Latency and Bytesent of the web objects are saved in proxy
server data store. The proxy server maintains cache hits of the
each web objects for a period of time and it is acting as
Frequency of the web objects. After a period of time, the cache
memory of proxy server would filled up. When new web
objects request come for placing in the cache of the proxy
server, then the replacement policy is applied for freeing space.
 The Sugeno type Fuzzy Inference System is employed here
to find a rank of web objects. This rank helps in replacement
policy. The rank is based on the input values of Byte Sent,
Frequency and Latency. The low valued rank web objects
would be removed first from cache of proxy server. If all the
web objects consider high rank then, arbitrarily any web
objects would removed for new coming web objects into cache
memory of Proxy Server.
3. FIS STRUCTURE
The Fuzzy Inference System (FIS) can be developed set of IF
Then rules and determine the membership function for the Fig 2: Bell shaped membership functions of inputs of
inputs. Here Sugeno type FIS is used. The sugeno based fuzzy FIS. The inputs are ByteSent, Frequency, and Latency.
Inference System can be explained as

21


Membership function Range from [0 1], that means the input Fig 3: FIS structure with different layers.
values must be between 0 and 1.
The Fig 3 FIS structure can be explained as Layer-1: Inputs
The inputs are the ratio of of the web objects Layer, Layer-2: Membership function for inputs, Layer-3:
different parameter. Rules, Layer-4: Output Membership function, Layer-5:
Here the inputs, Weighted sum output and Layer-6: Output.

4. IMPLEMENTATION
The response time, Delay of web objects can be obtained from
the Log file of the proxy server. The Log file can be look like
as in Fig 4.
Log file which maintains a history of web object requests and
And
use to analysis the user request.
The general contents of log files are
%a - Remote IP address
If then, the input would be considered as 1 %A - Local IP address
%b - Bytes sent, excluding HTTP headers, or '-' if zero
and the Maximum value would be updated with the Current %B - Bytes sent, excluding HTTP headers
Value . %h - Remote host name (or IP address)
Since it is Sugeno type FIS and it has 3 variables and each %H - Request protocol
variables having 2 regions “Small” and “Large”. So, the %l - Remote logical username
combination of rules is 23 i.e. 8. The outputs would be total 8 %m - Request method (GET, POST etc.)
numbers of constant type membership functions. %p - Local port on which this request was received
FIS Information: %q - Query string (prepended with a '?' if it exists)
Type: Sugeno %r - First line of the request (method and request URI)
Type of input Member Function: Bell function %s - HTTP status code of the response
No of Member Functions: 2 MFs for each input. %S - User session ID
Type of output Member Function: Constant. %t - Date and time, in Common Log Format
Number of fuzzy rules: 8 %u - Remote user that was authenticated (if any), else '-'
%U - Requested URL path
%v - Local server name
%T - Time taken to process the request, in seconds
%I - current request thread name (can compare later with stack
traces)
The pattern code is placed in configuration file of the proxy
web server. The proxy server would create a log file with the
required following data.
Out of the above here, only Response time (%T) acting as
Latency and Byte sent (%B) is retrieve for each web objects.
The Frequency for each web objects is maintained by proxy
server. The proxy server counts the number of cache object hits
for a particular period of time acting as Frequency for that web
object.

Fig 4: Log file Contents format.

22


When the cache memory is filled up with web objects, then the “Prefetching in Content Distribution Networks via Web
replacement algorithm would apply for freeing space for new Communities Identification and Outsourcing” World wide
coming web objects. The FIS would give a rank for each web Web (2008) 11:39–70; Springer.
objects. The rank range is [0 1] and the higher rank web objects [5]. Qiang Yang, Henry Haining Zhang, Ian T.Y. Li, and Ye
would give high priority to be cache and whereas the low Lu “Mining Web Logs to Improve Web Caching and
priority web objects would be removed early. Prefetching” School of Computing Science Simon Fraser
The ruler view is shown as in fig 5. University Burnaby, BC, Canada V5A 1S6 (qyang,
hzhangb, tlie, yel)@cs.sfu.ca
[6]. Areerat Songwattana “Mining Web logs for Prediction in
Prefetching and Caching” School of Engineering and
Technology, Asian Institute of Technology Rangsit,
Pathumthani, 12000, Thailand areerat@rangsit.rsu.ac.th,
sareerat@gmail.com
[7]. Qiang Yang and Haining Henry Zhang “Web-Log
Mining for Predictive Web Caching”.
[8]. Francesco Bonchi1, Fosca Giannotti2, Giuseppe
Manco3,Chiara Renso4,Mirco Nanni5,Dino
6 7
Pedreschi ,Salvatore Ruggieri “Data Mining for Intelligent
Web Caching”.1,2,3,4 CNUCE-CNR - Institute of Italian
National Research Council5,6,7 Department of Computer
Fig 5: Ruler view of Inputs with Outputs. Science, University of Pisa
[9]. Víctor J. Sosa Sosa1, Gabriel González S2., Leandro
The simulator is done with JDK 1.5 Netbeans IDE and Matlab Navarro3, “Building a Flexible Web Caching System” 1, 2
2009. JMatlink is used to Connection between JDK 1.5 and it is Centro Nacional de Investigación y Desarrollo
free software and can be downloaded from the link Tecnológico, Interior Internado Palmira S/N,
http://jmatlink.sourceforge.net/. The Fuzzy Logic Toolboxes in Cuernavaca,Morelos, México.Universitat Politècnica de
Matlab 2009 is used to design the Fuzzy Inference System. Catalunya (UPC)Jordi Girona, 1-3, D6-105, Campus
Nord,Barcelona, Spain.
5. CONCLUSIONS AND FUTURE WORK [10]. Sarina Sulaiman1, Siti Mariyam Shamsuddin2, Fadni
The membership functions of the Fuzzy Inference System need Forkan3, Ajith Abraham4 “Intelligent Web Caching Using
to refine its curves based on live data. So, live data required. Neuro computing and Particle Swarm Optimization
Next plan is to implement in practical place to check feasible or Algorithm”1,2,3Soft Computing Research Group, Faculty of
not. It would be better to use a trained Fuzzy Inference System Computer Science and Information System, Universiti
and Adaptive Neuro Fuzzy Inference System would be the Teknologi Malaysia, Skudai, Johor, Malaysia.4Centre for
solution for it. So, next plan is to modify the approach with Quantifiable Quality of Service in Communication
ANFIS (Adaptive Neuro Fuzzy Inference System) to refining Systems, Norwegian University of Science and
the membership function curves based on input to output data. Technology, Trondheim, Norway.
The propose paper is considering only Web Caching, but Web [11]. Michael Chau, and Hsinchun Chen “Incorporating Web
Prefetching would be another approach in which the anticipated Analysis Into Neural Networks: An Example in Hopfield
future web objects are placed in cache. Next plan is to propose Net Searching” IEEE
innovative Prefetching algorithm and integrate both the
technique Web Caching and Prefetching into the same Proxy [12]. S. V. Nagaraj “CACHING AND ITS APPLICATION”
Web Server. It is mentioned that the limitation of this approach Chapter:1 VARIOUS FLAVORS OF WEB CACHING,
required extra computation overhear for the proxy server and page 3-16
implementation complexity. [13]. Konstantinos Stamos, George Pallis, and Athena Vakali
6. REFERENCES “Integrating Caching Techniques on a Content Distribution
[1]. Wei-Guang Teng, Cheng-Yue Chang, and Ming-Syan Network” Konstantinos Stamos, George Pallis, and Athena
Chen “Integrating Web Caching and Web Prefetching in Vakali, Department of Informat Aristotle University of
Client-Side Proxies” Fellow, IEEE. Thessaloniki, 54124, Thessaloniki, Greece.
[2]. Jaeeun Jeon, Gunhoon Lee, Ki Dong Lee and Byoungchul [14]. S. V. Nagaraj “CACHING AND ITS APPLICATION”
Ahn, “An Adaptive Prefetching Method for Web Caches” Chapter:11 PREFETCHING, page 105-122
Yeungnam University, School of Electrical Engineering [15]. Waleed Ali & Siti Mariyam Shamsuddin “Neuro-Fuzzy
and Computer Science. System in Web Client-side Caching” Faculty of Computer
[3]. L. Y. CAO and M. T. OZSU “Evaluation of Strong Science & Information System, University Technology of
Consistency Web Caching Techniques” University of Malaysia
Waterloo, School of Computer Science, Waterloo, ON, [16]. Josef Schmidbauer and Hilmar Linder “Utilizing
Canada N2L 3G1. Layered Multicast for Web Caching” Department of
[4]. Antonis Sidiropoulos, George Pallis, Dimitrios Katsaros, Scientific Computing, Paris-Lodron University of Salzburg,
Konstantinos Stamos,Athena Vakali, Yannis Manolopoulos Salzburg, Austria

23

An Optimization Technique of Web Caching using Fuzzy Inference System

More Related Content

What's hot (18)

Similar to An Optimization Technique of Web Caching using Fuzzy Inference System (20)

An Optimization Technique of Web Caching using Fuzzy Inference System