1. Kalyan Boggavarapu Lehigh University
A Protocol-Independent Technique for Eliminating Redundant Network Traffic
Neil T.Spring and David Wetherall
Computer Science and Engineering,
University of Washington
Presented by Kalyan Boggavarapu, Lehigh University
11/16/2018
Kalyan Boggavarapu Lehigh University

2. Kalyan Boggavarapu Lehigh University
Introduction
11/16/2018
Kalyan Boggavarapu Lehigh University

3. Kalyan Boggavarapu Lehigh University
Aim
To eliminate redundant transfers
Currently highest cacheabitly obtained is 45% by squid. ( to improve that)
Origins of uncached content
Dynamically generated or personalized
Mirrored on a different server
Named by a different URL
Delivered using a new or unsupported protocol
Updated static content
Access counted for advertising revenue
11/16/2018
Kalyan Boggavarapu Lehigh University

4. Kalyan Boggavarapu Lehigh University
Algorithm
11/16/2018
Kalyan Boggavarapu Lehigh University

5. Kalyan Boggavarapu Lehigh University
Rabin Fingerprint
Rabin fingerprint for a sequence of bytes t1,t2,t3,…tß
ß = Length of the string
M = 260
P =
Make a table of ti x ß values for fast access (256 entries)
RF/FP are calculated for consecutive bytes of length of ß
Eg:
1,2,3,4,5,6 bytes,where ß=3
1,2,3
2,3,4
3,4,5
11/16/2018
Kalyan Boggavarapu Lehigh University

6. Kalyan Boggavarapu Lehigh University
Algorithm
Hold recent packets in the cache
Maintain a FP index of the first bytes of the packets
Check for each incoming packets
Follow LRU for cache replacement
11/16/2018
Kalyan Boggavarapu Lehigh University

7. Kalyan Boggavarapu Lehigh University
Architecture
11/16/2018
Kalyan Boggavarapu Lehigh University

8. Shared Cache Architecture
Not implemented
Inconsistency is detected by New FPs
Protocol independent
Web Caching + Delta transmission
Bandwidth constrained link
Packets = tokens + encoded
11/16/2018
Kalyan Boggavarapu Lehigh University

9. Kalyan Boggavarapu Lehigh University
Implementation
11/16/2018
Kalyan Boggavarapu Lehigh University

10. Kalyan Boggavarapu Lehigh University
Parameters Selection
Selected values are β=64 bytes γ = 5; Throughput is inversly proportional to β, γ; Cache Allocation = 40% for Fp Index, 60% cache
11/16/2018
Kalyan Boggavarapu Lehigh University

11. Kalyan Boggavarapu Lehigh University
Amount of Redundancy
Incoming is redundancy can be captured by increasing the cache size
Outgoing data redundancy has limit
11/16/2018
Kalyan Boggavarapu Lehigh University

12. Locality of Redundancy
Same server, lot of redundancy, therefore we can apply end-end solutions
Name based proxy caching would fail to capture this amount of redundancy
11/16/2018
Kalyan Boggavarapu Lehigh University

13. Kalyan Boggavarapu Lehigh University
Traffic Analysis
11/16/2018
Kalyan Boggavarapu Lehigh University

14. Kalyan Boggavarapu Lehigh University
Link Utilization
Less utilization => more Byte savings
(cache is divided among less number of users)
More utilization => less Byte savings
( can be improved by increasing the cache size)
11/16/2018
Kalyan Boggavarapu Lehigh University

15. Redundant incoming traffic
High
11/16/2018
Kalyan Boggavarapu Lehigh University

16. Kalyan Boggavarapu Lehigh University
Summary
Best
11/16/2018
Kalyan Boggavarapu Lehigh University
To eliminate requests
Long Term repitition

17. Kalyan Boggavarapu Lehigh University
Related Work
11/16/2018
Kalyan Boggavarapu Lehigh University

18. Duplicate Suppression
Similar to this method
β is larger or equal to the packet size
Calculate similarity between larger regions
Achieved suppression is less
Delta Encoding
Send the differences in the page
11/16/2018
Kalyan Boggavarapu Lehigh University

19. Kalyan Boggavarapu Lehigh University
Conclusion
Redundancy suppression > Duplicate Suppression
Redundancy suppression > Delta encoding
Protocol independent
Useful to suppress 60% of the incoming traffic and 30% of the outgoing traffic
11/16/2018
Kalyan Boggavarapu Lehigh University

20. Kalyan Boggavarapu Lehigh University
Questions !
11/16/2018
Kalyan Boggavarapu Lehigh University