Anda di halaman 1dari 28

c 



  
 
    
Aniket Mahanti, Anirban Mahanti, and
Carey Williamson

University of Calgary, Canada


   

    

          

 

ï Motivation
ï ADF framework
ï bjectives

 

ï Web caching proxies are an effective


means of reducing network traffic
ï Web caches are widely deployed by ISPs
ï Caches improve performance by
exploiting workload characteristics
such as locality of reference
ï Workload characterisation of locality
structure can provide insight into the
design and performance of the Web
Ä
i  

ï ocality characteristics can be used by


caching policies when making decisions
to evict or retain documents in the cache
ï Most prior Web caching work focused on
analysing Web streams in | |
ï Fonseca et al. 2005] proposed a
³system level´ view called the ADF
framework for analysing transformations
of a Web reference stream
K
@ 

!"  #

@@
 
  
 



Reference: R. Fonseca et al. (2005), ocality in a Web of Streams, In: V|| 


 V , K (1): 2² . 5
 
 $% 

ï Study locality properties in Web request


streams using the ADF framework:
ËWhat impact do locality characteristics
have on caching performance?

ËWhat are the locality characteristics of


Web request streams after the
aggregation of filtered streams?

†
]!


ï Flow of requests
ï ocality of reference
   & 

Image reproduced from: R. Fonseca et al. (200Ä), ocality in a Web of Streams, Technical Report, Department of Computer Science, Boston University.
c  


ï Popularity: An object is simply more


popular than other objects
««.XABXXCXDXXXEFXX««.

ï Temporal locality: References to an


object occur in a correlated manner
««.AAHIJAAAU PJKAA««.

 
i 
 ' 


 i 


ï Document hit ratio: Percentage of total


requests satisfied by Web proxy cache
ï Byte hit ratio: Percentage of total byte volume
of data satisfied by Web proxy cache

ï Cumulative reference measure: Fraction of


total requests accounted for by the top 10%
of the most popular documents
ï Inter-reference measure: Probability of
referencing document again within
intervening requests (e.g., )
11

i  

ï Model description
ï Simulation results

i  

Filtered stream (misses)

Filtering

Input stream



!   
 

ï WebTraff: synthetic Web proxy workloads
ï Two traces differing only in temporal locality
± Trace1 (weak) and Trace 2 (strong)
ï Trace characteristics:
± 1.5 million requests
± K 5,000 unique documents
± 1K GB total bytes of Web content
ï Cache replacement policies:
± RU, FU-Aging, GDS, RAND, FIF
ï Cache size:
± 1 MB ± 1† GB 1K
 

 () *+

r  : Weak temporal locality


 , 15
 

 ( *+

r  : Strong temporal locality


 ,

 

 (* *+

r  : Weak temporal locality r  : Strong temporal locality


 ,

17
 
 

  

r  : Weak temporal locality r  : Strong temporal locality


 
 i  



 & 

( 
  + 1
 
c  

 

r  : Weak temporal locality r  : Strong temporal locality



-

 i  



 & 

( 
  + 1 
@
 i  

ï Model description
ï Simulation results
@
 i  

Aggregated
filtered stream

Filtered stream
Filtering

Input stream

21
  i  
 

ï Two-level hierarchal web proxy configuration

ï Aggregated streams from:


± N = 1, 2, K, child proxies
ï Caching policy:
± RU at child proxies
ï Cache size:
± 1 MB ± 25† MB
ï Degree of overlap:
± No overlap, partial overlap

22
 
 

  

ß    |  


 
 i  

(
 
 +

 
c  

  

ß    |  



-

 i  

(
 
 +
2K
ß $
 
c 

ï Temporal locality 2   with |  | ß


ï Phenomenon consistent over various cache
sizes and degree of temporal locality
ï Design of the no overlap scenario
N=2
Child Proxy 1: 1A1,1U1 ,1U2,««.,1U50,1A1
Child Proxy 2: 2A1,2U1 ,2U2 ,««.,2U50,2A1
Aggregated filtered stream: 1A1,2A1,1U1,2U1,«««, 1U50,2U50,1A1,2A1
,

ï New stream has twice as many documents


between 1A1 and 2A1
25

$
 
c 

ï Temporal locality |  with |  | ß


ï Due to 50% overlap among all traces
N=K
Child Proxy 1: A, B,1U1 ,««.,1U50, A, B
Child Proxy 2: A, B,2U1 ,««.,2U50, A, B
Child Proxy Ä: A, B,ÄU1 ,««.,ÄU50, A, B
Child Proxy K: A, B,KU1 ,««.,KU50, A, B
Aggregated filtered stream: A, A, A, A, B, B, B, B,1U1 ,2U1 ,ÄU1 ,KU1
,««.,1U50,2U50,ÄU50 ,KU50, A, A, A, A, B, B, B, B
,
ï References of A from other proxies clustered

   
   

ï Caching policies should exploit   


 | and  | of documents
ï RU and FIF exploit temporal locality
ï GDS insensitive to changes in temporal locality
ï Structural change in temporal locality for
aggregated streams depends on the 2 

  in the workloads
ï These results imply limited advantages of
using caching hierarchies