Transcript Slides

Using Proxy Cache Relocation to Accelerate
Web Browsing in Wireless/Mobile Comm.
Authors:
Stathes Hadjiefthymiades and Lazaros Merakos
Dept. of Informatics and Telecommunication – Uni. of Athens, Greece
Proceedings of The Tenth International World Wide Web
Conference, May 1-5, 2001, Hong Kong.
Contents

Web caching in wireless environment
 “Moving” cache architecture
 Cache relocation scheme
 Path prediction algorithm
 Performance evaluation
 Simulation results
 Conclusion
Web caching in wireless environment

MobiScape/Web Expess (1995/96)
– Support Station (SS) is a gateway of Mobile
Host (MH)
– Both use proxy cache
– Data is compressed
– No changes in browsers, servers
– SS must be reconstructed each time MH
changes cell
“Moving” cache architecture

Components: base stations (BS), mobile terminal,
fixed terminal, routers.
 Wireless cells: hexagonal shape, cover the entire
surface.
 User profiles: stored in home network, can be
queried and forwarded using inter-network
signaling.
 Path prediction algo. : invoked after entering new
cell, may be stored at home network
Cache relocation scheme

A relocation process has these steps:
– Determine_target[MT_ID,BS_ID]: MT to Home
– Path prediction algorithm: Home
– [MT_ID, Target_BSs, HO_Probabilities]: Home to
MT
– Cache compression: MT
– MT_Cache[MT_ID,Cache]: MT- new BSs
– Cache Decompression: New BSs
– Handover: MT
– Feedback[MT_ID,BS_ID]: New BS to Home
– Clear_cache[MT_ID]: Home to unused BSs
Cache relocation scheme

Move 100% to best guessed new BS, 70%
to 2nd best guessed BS, 30% to other BSs.
Path prediction algorithm

Based on learning automaton (an AI machine
learning technique).
 Learning automata:
– Finite state adaptive systems that interact continuously
with an environment.
– Learn to adapt through a trial-error response process.
– Input  Responses  Evaluate response  Feedback
 Improve behavior.
– Robust but not very efficient learners. Easy to
implement.
Path prediction algorithm

Main steps:
– Receive prediction request.
– Lookup matrix, send responses.
– Receive feedback, update matrix

Matrix maintenance
Performance evaluation setup

WWW traffic modeling: figure 9.
 Cell residence time: time spent in current cell.
This time is short if user travels very rapidly (in
vehicle), it is long if user travels slowly (walking).
 Path prediction programmed in Prolog.
 Cache relocation scheme programmed in Visual
C++. Metrics: avg. delay, # of interrupted
connection, % of interrupted conn., hit rate, # of
items used by MT in the new BS after handover.
Simulation results


Path prediction algorithm
Cache relocation scheme
Conclusion

Introduce a cache relocation and path
prediction scheme for WWW browsing in
wireless/mobile environment.
 More robust learners in path prediction
algorithm could be use.
Comments




Relocating data: didn’t mention how second best guesses
share data; how many second best guesses in general.
Path prediction: could be run from current BS without
contacting home network.
Performance evaluation: didn’t compare with existing
techniques. Didn’t study wasted bandwidth used for
transfer data in incorrect predictions.
Contributions are not very clear since this technique adopts
many things from existing techniques (architecture from
MobileSpace, prediction algorithm from AI).
Discussion