290 likes | 415 Views
PTC: Proxies that Transcode and Cache in Heterogeneous Web Client Environments. Aameek Singh, Abhishek Trivedi, Krithi Ramamritham (IIT Bombay) AND Prashant Shenoy (University of Massachusetts, Amherst) Web Information Systems Engineering-02. Outline …. Introduction Transcoding Proxies
E N D
PTC: Proxies that Transcode and Cache in Heterogeneous Web Client Environments Aameek Singh, Abhishek Trivedi, KrithiRamamritham (IIT Bombay) AND Prashant Shenoy (University of Massachusetts, Amherst) Web Information Systems Engineering-02
Outline … • Introduction • Transcoding Proxies • Caching Policies • Cache Replacement • Implementation and Results • Conclusions and Future Work
Introduction • Diverse client devices • Differ in hardware, software and network connectivity • Each client requires specific content for efficient rendering • Different versions of same data • Image Resolution/Quality • Level of text summarization
Transcoding • Large number of client-specific versions • Conversion of one data version to another • Decreasing Image Quality (JPEG quality level) - “convert” utility in Linux • Summarizing text- Copernicus Summarizer
Possible Solutions • Server keeps all versions • Server transcodes on-the-fly • Intermediate Proxy Diversity! Resources! • More Control Appropriate! • Utility from an ISP
Caching • Proxy Caching • New Issues • Multiple Versions of data • Transcoding possibilities in the cache • Cache Replacement
Caching • Full Hit • Desired version available • Partial Hit • Higher version available • Secondary Hit • Lower version available • Miss • No version available
Partial Hit • Two possibilities • Download desired version from server • Transcode higher version locally • Factors influencing decision • Transcoding Complexity • Proxy-server network connection • Load on proxy
Proposed Strategy • Predict time to transcode (tAB) • Version before & after transcoding • Size of the object • Load on proxy • Predict time to download (tS) • Size of object • Network speed • Load on proxy
Adaptive Policies • IP Based • Filter based on IP address of server • Min-Min • Assume next transfer to be quickest • Multiple Linear Regression • Predict based on a linear model • Complex Policies
Proposed Strategy • Heuristic if M*tS < tAB then Download else Transcode locally • M – Level of conservativeness
Secondary Hit • Proxy has a lower version • Need to integrate personalization • Profiling at proxy • Permanent Profiles • Session Characteristics • More utilities can be added
Cache Replacement • Two kinds of utilities with each object • Reference utility • Transcoding utility • Cost savings important • WATCHMAN – Caching of Query Results
Cache Replacement • Profit Metric of Oi – λc/s • λ - Average Reference Rate • c - Cost of obtaining Oi • s - Size of object • Extension • Include Transcoding utility • Add a term γb/s for each version, Oi can be transcoded into.
Cache Replacement • Evaluation of parameters • λ = K/(t-tk) • c = min(ts, tVjVi), j = min(i+1,..n) • b = min(ts – tViVj, tVkVj – tViVj), k = min(i+1,..n) • Maintenance of parameters
Related Work • Chang, Chen – ICDE 2002 • Weighted Transcoding Graphs • Constant Server Download Rate • Constant Transcoding Rate • Need Smarter policies
Implementation and Results • Traces • 3000 JPG files • 5 concurrent clients with 3600 requests following Zipf’s Law (alpha=0.5) • Request stream consisting of 1200 Partial Hits • Local and Remote Source Servers
Results • Local Source Server
Results • Remote Source Server
Breakup of Performance • Local Source Server
Breakup of Performance • Remote Source Server
Conclusions and Future Work • Caching is essential and newer strategies are required • Adaptive policies are essential • Even simple policies will work • More policies and real-life client simulations are required • Scope of data objects should be increased