Download presentation
Presentation is loading. Please wait.
Published byMichael Simpson Modified over 9 years ago
1
System Software Lab 1 Enhancement and Validation of Squid ’ s Cache Replacement Policy John Delley Martin Arlitt Stephane Perret WCW99 김 재 섭 EECS System Software Lab.
2
System Software Lab 2 Contents Introduction Web Cache Architecture Explicit proxies & Transparent proxies Role of the Replacement Policy Related Work LRU, LFU, LFU-Aging, GDS, GDS-F Replacement Policy Implementation Simulation Simulation Results Conclusion & Critiques
3
System Software Lab 3 Introduction Web Cache Architecture Explicit Cache Transparent Cache
4
System Software Lab 4 Role of the Replacement Policy Determine future popularity of documents Use Resource in the most advantageous way WebProxy Workload Characterization Tremendous breath and turnover in popular object Measures of Efficiency Hit rate Byte hit rate Bytes returned directly from the cache / Total byte accessed CPU or IO system utilization Average object retrieval latency Inversely proportional to object hit rate Hard to estimate
5
System Software Lab 5 Related Work LRU Temporal locality Single linked list Removal of object =>O(1) Update takes constant time LFU Heap Cache pollution LFU-Aging GDS(Greedy Dual Size) Consider account size and cost function for retrieving object GDS-F
6
System Software Lab 6 Replacement Policy Implementation LFU with Dynamic Aging Variant of LFU Accommodate shift in the set of popular object Insertion & Modification Reference count + cache age No needs for parameterization GDS-Hits Variant of GDS Optimized for popular smaller objects Reference count/size + cache age
7
System Software Lab 7 Simulation Implementation experiences Squid 1 LRU is implemented scanning a set of hash buckets and qsort Not precise Squid 2 Heap based 36,000 unique objects 8M memory 100M Disk
8
System Software Lab 8 Simulation Results (Squid 1)
9
System Software Lab 9 Simulation Results (Squid 1)
10
System Software Lab 10 Simulation Results (Squid 2)
11
System Software Lab 11 Simulation Results (Squid 2)
12
System Software Lab 12 Conclusion & Critiques Conclusion Accommodate shift in the set of popular object Two frequency-based cache replacement policies Critique In LFU-DA reference count can increase infinitely GDS-Hits Only concentrate on hit rate by caching small object What are the benefits for user? Ideas are too trivial and its simulations are also unreliable. What ’ s difference between Squid 1 and Squid 2 ?
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.