160 likes | 266 Vues
The seminar presented by Yutao Zhong discusses the critical issues affecting web performance, including network congestion and server overloading, which lead to increased user-perceived latency. To combat these challenges, web caching emerges as a robust solution, storing popular objects closer to users via client-oriented, server-oriented, and proxy caching strategies. The advantages of web caching include reduced bandwidth usage, lighter server loads, and diminished latency. The seminar also addresses design issues like cache misses and stale data, outlining various cache replacement and prefetching strategies to ensure data coherence and optimal performance.
E N D
World Wide Web Caching CS457 Seminar Yutao Zhong 11/13/2001
Problems • Network congestion • Server overloading • User-perceived latency
client Web server Web cache Effective solution: Web caching • Storing popular objects at locations close to the clients • Different deployments • client-oriented, server-oriented, proxy
Advantages of Web caching • Reduces bandwidth usage • Lightens Web server loads • Lessens user-perceived latency
client Web client cache router client Proxy cache (I) • Standalone cache
client Web client router client cache cache cache Proxy cache (II) • Transparent cache • Router level
Web client L4 switch client router client cache cache cache Proxy cache (III) • Transparent cache • Switch level
Design issues • Latency caused by cache miss (cache replacement, prefetching,…) • Stale data (cache coherence) • Dynamic data (dynamic cache)
Cache replacement • Traditional policies • LRU, LFU, FIFO • Key-based policies • LRU-MIN, LRU-Threshold, HyperG • Function-based policies • Bolot/Hoschka, LNC-R (least normalized cost-replacement)
Cache prefetching • Prediction • server-based vs. user-based • Scenarios • poll vs. push • Target • content vs. means • preresolving, preconnecting, prewarming
Cache coherence:deals with stale data • Data validation • HTTP assistance • Conditional GET (IF-Modified-Since) • Date: date • Last-Modified: date • Expires: date • Pragma: no-cache • Cache-control: max-age
Cache coherence:implementing mechanisms • Client polling • Server invalidation • Adaptive TTL • Piggyback invalidation
Dynamic data caching • Active caching • cache computation • CPU cost • Web server accelerator
Typical systems • CERN httpd: 1994 • Harvest (Squid): 1995 • Akamai
Summary • Effective solution to Internet scalability problem • Issues and techniques • Open problems • cache routing, dynamic data caching, security...