Integrating Web Caching and Web Prefetching in ClientSide Proxies

1
Integrating Web Caching and Web Prefetching in
Client-Side Proxies

• Wei-Guang Teng, Cheng-Yue Chang, and
• Ming-Syan Chen, Fellow, IEEE

IEEE TRANSACTIONS ON PARALLEL AND DISTRIBUTED
SYSTEM, VOL. 16, NO. 5, MAY 2005
Present Yi-Wei Ting
2
Outline

• Introduction Motivation
• Web caching and prefetching
• System environment
• System model and prefetching rule
• Cache replacement algorithm
• Normalized profit function
• Algorithm IWCP (Integration of Web Caching and
  Prefetching)
• Performance analysis
• Simulation model
• Experimental results
• Conclusion

3
Introduction

• Web caching
• Exploits the temporal locality.
• .Clients browser caching, client-side proxy
  caching, and server-side proxy caching.
• Web prefetching
• Utilizes the spatial locality of the Web objects.

4
Motivation

• Web server may send all possible prefetching
  hints with various levels of confidences to the
  proxy, a proxy will prefetch every implied object
  into its cache.
• On the contrary, it the prefetching cotrol is
  over strict, a proxy will tend to discard some
  beneficial hints provided by the Web server, thus
  whittling down the advantage of Web prefetching.

5
Example 1
Reference rate A, B, C, D, E ? 3,1,1,4,2. Cache
capacity 2
Cache hit
Without prefetching
C
A
D
A
D
E
D
E
A
D
B
D is replaced by C
With prefetching
C
A
D
A
D
E
D
E
A
D
B
Assume the prefetching rule (B?C) holds with the
confidence of 100 percent.
6
Example 2
Client 1
Client 2
Client 2
A
B
A
B
0.6
Client 1
Client 3
Client 3
C
D
E
C
D
E
0.7
0.8
0.6
0.8
F
G
H
F
G
H
7
System environment
The system model of Web prefetching and Web
caching.
8
Prefetching rules

• Let O o1,o2,,on be a set of Web objects in a
  Web server and D be a set of users access
  sequences, where each access sequence is a set of
  objects in O ordered by time.
• Definition 1. A prefetching rule is an
  implication of the form o1,,oi ? oi1, where
  o1,,oi,oi1 eO, o1,,oi,oi1 is an access
  sequence in D, and c is the confidence of the
  prefetching rule.

c
9
Prefetching rules

• Definition 2.
• The confidence c of the prefetching rule o1,,oi
  ? oi1 is a conditional probability of
  P(o1,,oi,oi1) / P(o1,,oi),
• where
• P(o1,,oi,oi1 ) is the probability that the
  sequence o1,,oi,oi1 is contained in an access
  sequence in D.
• P(o1,,oi), is the probability that the sequence
  o1,,oi is contained in an access sequence in D.

c
10
Prefetching rules

• Definition 3.
• An object oi1 is referred to as an implied
  object if and only if the prefetching rule
  o1,,oi ? oi1 is triggered by some client who
  has already referenced the objects o1,,oi in its
  precedent requests.
• Otherwise, the object oi1 is called a
  nonimplied object.

c
11
Normalized profit function

• We shall first calculate the expected number of
  references to object oi within the time window w.
• Nonimplied object
• ?iw
• Implied object, multiimplied object
• ?iw ci,j

12
Normalized profit function

• Property 1. Let p1, p2,,pn denote the
  probabilities that the object oi will be
  referenced once within the time window w by
  clients 1,2,,n, respectively.
• The expected number of references to object
  oi within the time interval w is then equal to p1
  p2 pn.

13
Normalized profit function

• Theorem 1. Let R denote the set of the
  prefetching rules that have been triggered to
  prefetch oi. Let ni,j denote the number of the
  clients that triggered the rule ri,j in R.
• Then, the expected number of references to
  object oi with in the time window w is equal to
• ?iw Sri,jeR ni,j ci,j

14
Normalized profit function

• Theorem 2. The profit from caching object oi is
  equal to
• Pf (oi) (?iw Sri,jeR ni,j ci,j ) di
  uiw vi
• Where the expression uiw vi calculates the
  extra delay incurred by validation checks for
  object oi within the time window w.

15
Normalized profit function
IWCP (2005)

• (?iw Sri,jeR ni,j ci,j )
  di uiw vi

Pf ( oi )
si
16
An example to illustrate the linked list for the
purpose of tracking the prefetching rules
Client 1
Client 2
Client 3
C
D
E
0.7
0.6
0.8
F
G
H
c
Expire time
Node id
G 1.3
1 0.7 5
2 0.6 7
H 0.8
3 0.8 6
17
An illustrative example for Algorithm IWCP
18

• Algorithm IWCP (oi)
• 1. if (oi ! C) // object not found in proxy
• 2
• 3. fetch oi from the origin server and then send
  oi to the client
• 4. insert oi into C and extract the piggybacked
  hinted objects
• into H
• 5. calculate PFN(oi)
• 6.
• 7. else // object found in proxy
• 8. send oi to the client
• 9. search Li to delete the node which can be
  satisfied by the
• request for oi
• 10. update PFN(oi)
• 11. fetch from the origin server those hinted
  objects triggered
• by requesting oi

19

• 12. extract the hinted objects into H
• 13.
• 14. for (each hinted object oh e H)
• 15. if (oh e C)
• 16. insert (client chj timer(now w)) into
  Lh
• 17. update PFN(oh)
• 18. else
• 19. insert oh into C
• 20. insert (client chj timer(now w)) into
  Lh
• 21. calculate PFN(oh)
• 22.
• 23.
• 24. BuildHeap (C)
• 25. while (Snow gt cacheCapacity)
• 26. evict the first object oj from C
• 27. Snow Snow - sj
• 28.

20
Normalized profit function
IWCP (2005)

• (?iw Sri,jeR ni,j ci,j )
  di uiw vi

Pf ( oi )
si
LNC-R-W3 (1999)
(ri di ui vi)
Pf ( oi )
si
21
Parameters of the Simulation Model
22
Delay saving ratios under various cache capacities
23
Conclusion

• Developed Algorithm IWCP by integraing Web
  caching and Web prefetching in client-side
  proxies.
• Formulated a normalized profit function to
  evaluate the profit from caching an object.