Improvement of Web System Performance Using Data Mining

1
Improvement of Web System Performance Using
Data Mining
2
CONTENTS

• INTRODUCTION
• CONCEPTS
• SYSTEM ARCHITECTURE
• FUTURE WORKS
• REFERENCES

3
INTRODUCTION

• WEBs popularity HIGH LATENCY
• A solution to decrease the latency is CACHING.
• USING ONLY THE CONCEPT OF LOCALITY.
• There may be some associativities among users
  request to a web server.
• The enhanced solution is
• CACHING COMPLEMENTED BY THE PREDICTION OF
  USERS BEHAVIOURS IN THE REQUEST STREAM

4
ASSOCIATION RULE AN EXAMPLE (1)

• One student took some courses
• O/S P.M means that a student who takes
  O/S also takes P.M.

Student Courses
A O/S, Math, P.M
B O/S, Pattern
C O/S, P.M
D Math, Embeded, Model
5
ASSOCIATION RULE AN EXAMPLE (2)

• SUPPORT of O/S P.M
• 50
• Because A,C transaction among four transactions
  have this rule
• CONFIDENCE of O/S P.M
• 66.7
• Because A,C students take PM among three students
  who take O/S.

6
Pre-Fetching

• Using previous association rules.
• The mechanism of deducing forth-coming page
  accesses of a client.
• Based on access logs of webserver.
• There are some kinds of pre-fetching algorithms.

7
PEC Algorithm

• Prodedure Prefetch(Array R, int M, float
  maxSize)
• //h,b are sequence of document ids
• begin
• prefetchSeq 0
• foreach rule h ? b such that h ? R
• foreach d ? b such that d.size lt maxSize
• prefetchSeq prefetchSeq ? d
• endfor
• endfor
• sort document in prefetchSeq in decreasing order
  of the confidence of the corresponding rule and
  keep the first M ones.
• return prefetchSeq
• end

8
Caching Procedure

• Procedure PECache (Array R, Document d)
• Begin
• R R ? d
• if not ( d in Weighing Room or d in Waiting
  Room)
• put d at head of the LRU list of the Weighing-
  Room
• prefetchSeq Prefetch(R, M, maxSize)
• foreach p in prefetchSeq
• append p at the end of Waiting Room que
• endfor
• elseif d in Waiting Room
• remove d from Waiting Room
• put d at head of the LRU list of the Weighing
  Room
• else if d in Weighing Room
• put d at head of the LRU list of the Weighing
  Room
• else if
• end

9
System Architecture
Enhanced Webserver System
Web Server log files
I N T E R N E T
document request
logging
request
WEB CLIENT
read
response
Request Service Module
response document
Prediction Engine
prediction
lt Predictoin-enabled Web Server gt
10
Future works

• Analysis web caching policies of existing web
  server, such as apache, Jigsaw.
• Implementation of Prediciton Engine.
• Applying this prediction engine into existing web
  server.
• Research about the method of testing the
  performance of the predictivity and the user
  percieved latency.
• Performance Test.
• Prepare final presentation.

11
References

• 1 Alexandros Nanopoulos, Dimitrios Katsaros,
  and Yannis Manolpoulos Exploiting Web Log
  Mining for Web Cache Enhancement.
• 2 Discrete mathematics and its applications,
  4th ed.
• 3 Jaideep Srivastava, Robert Colley, Mukund
  Deshapnde, Pang-Ning Tan, Web Usage Mining
  Discovery and Applications of Usage Patterns from
  Web Data, ACM SIGKDD, Jan 2000.

12

• END