Building a Large and Efficient Hybrid Peer-to-Peer Internet Caching System

1
Building a Large and Efficient Hybrid
Peer-to-Peer Internet Caching System

• Li Xiao
• Xiaodong Zhang
• Artur Andrzejak
• Songqing Chen

2
Overview

• Web contents become more diverse
• Multiple levels of caching
• Document duplication among different levels
• Underutilized network between clients
• Share cache contents
• P2P web caching management scheme
• Improve scalability

3
Typical web caching

• Browser cache / Proxy cache
• Enlarge cache size
• Documents exist in browser cache but have been
  replaced in the proxy
• Decrease of proxy hit ratio
• Each document is replicated in local cache

4
Limitations due to replication

• Overhead
• Invalidation
• Broadcast
• Waste storage space
• Close the speed gap between local and remote
  accesses
• Transfer among clients

5
Browser aware caching

• Share cache contents
• Reduce document replication
• Utilize web contents / network bandwidth
• Hybrid P2P system
• Issues
• Data integrity
• Privacy

6
Trends in web caching

• Dynamic pages gt non-cachable
• Increase of cache size
• Can not keep up
• Hit ratios are reduced

7
Trends in web caching (contd)
8
Duplication in web caching

• Types of data
• Requested by a single client
• Requested by a multiple clients
• Intrasharing / Intersharing ratio

9
Architecture

• Proxy server keep track of client cache contents
• Miss -gt try to find in another clients cache
• Forward data / copy in proxy cache

10
Architecture (contd)

• Local hit
• use it )
• Proxy hit
• (Access counter from client)
• Store in client if counter above threshold
• Proxys browser index file
• Access counter (global and specific for client)
• Move it to proxy / cache in another client
• Global miss
• Fetch document / store only in client

11
Data structures

• Browser index file in proxy
• Id of client / URL
• Client initiated invalidation
• Browser data
• remote accesses
• Proxy data
• remote accesses
• Duplicate document?

12
Evaluation Cache size

• Proxy memory for each client 0.04 - 0.08 Mb
• Proxy disc space for each client 7 11 MB
• (

13
Evaluation Proxy cache size
14
Evaluation Browser cache size
15
Evaluation Replacement threshold
16
Evaluation Scaling clients
17
Evaluation Latency

• 21 reduction compared to basic scheme
• 56 reduction compared to typical scheme

18
Reliability and Privacy

• Data integrity
• Clients can modify files
• Digital watermark
• Privacy
• Do not reveal client requests / hide identities
• Proxy acts as anonymizer
• Receive / forward requests (where is p2p?)

19
Implementation

• Client daemon
• Mozilla
• Browser aware proxy server
• Squid
• For average size of 8 Kb, 11ms overhead
• Fetch from web server -gt 50ms

20
Conclusion

• Hit ratio decrease, significant duplication
• Browser aware caching, share cache contents
• Data integrity and privacy issues