Hint-based Acceleration of Web Proxy Cache

Daniela Rosu Arun Iyengar Daniel Dias IBM T.J.Watson Research Center

Unversity of Yuan Ze,Syslab Mike Tien mike@syslab.cse.yzu.edu.tw

Outline 1.Accelerated Web Proxy 2.Hint Representation 3.Hint Consistency 4.Comparison 5.Proxy Accelerator Cache 6.Throughput Improvements 7.Impact of Hint Management 8.Summary 8.Conclusion

1.Accelerated Web Proxy To speed up proxy caches by

reducing their cache-miss overhead. Extended content-based router

called a proxy accelerator(PA). With embedded O.S optimized for

communication.

Accelerated Web Proxy(cont.)

Accelerated Web Proxy(cont.)

2.Hint Representation Accurate–information directory scheme (Hash Table) Approximate-information based on Bloom Filters --bitmap(hint space)

PA(entries,hash table)

WP(collision counter)

Hint Representation(cont.)

--independently & interleaved

3.Hint Consistency Protocol ensure that the information at the PA

site reflects the content of the WP Protocol choices --entity that identifies the updates --protocol initiator --rate of update messages --exchanged information Previously proposed --WP-only update --eager hint register

4.Comparison

5.Proxy Accelerator Cache PA should maximize the hit ratio --size , GDS(1) WP should minimize the Network-

related costs

6.Throughput Improvements(PA improve WP throughput)

Throughput Improvements(cont.)(PA cpu speed affect improvement)

Throughput Improvements(cont.)(PA-level cache improve throughput)

7.Impact of Hint Management

Impact of Hint Management(cont.)

Impact of Hint Management(cont.)

Impact of Hint Management(cont.)

Impact of Hint Management(cont.)

8.Summary Bloom Filter-based scheme exhibits lower

computation and communication overhead WP meta-data may reduce its memory hit ratio Hint look-up overheads are small with respect

to the typical request processing overhead Eagar hint registration reduces update-related

overheads and prevents hit ratio reductions caused by update delay

False hit ratio is more significantly affected by hint representation than by update period

9.Conclusion This miss ratios at proxy caches are often

relatively high,we have developed a method to improve the performance of a cluster-based Web proxy by shifting some of its cache miss-related functionality to be executed on a proxy accelerator

Eagar registration is a “must” and that the Bloom Filter-based scheme is more appropriate than the directory scheme for large WP clusters or when PA and WP nodes have comparable power