1. Proxy Caching the Estimates Page Load Delays Roland P. Wooster and Marc Abrams Network Research Group, Computer Science Department, Virginia Tech 元智大學 資訊工程所 系統實驗室 陳桂慧 1999.05.11

2. Outline • Related results • Removal algorithms • Latency Estimation Algorithm (LAT) • Hybird Algorithm (HYB) • Results of experiment

3. A removal algorithm can optimize one of several measures: • minimize the number of bytes that travel over the Internet • minimize the number of hits to popular Web servers; or • minimize the time that end users wait for a document to load.

4. Related Results • The evidence in Williams et al [WILL96] is that proxy caches can • minimize server loads by using SIZE as a removal algorithm, • minimize wasted network bandwidth by using LFU, and achieve hit rates in the 20-50% range.

5. Latency Estimation Algorithm (LAT) • clatj = (1-ALPHA) clatj + ALPHA sclat • cbwj = (1-ALPHA) cbwj + ALPHA scbw. • Clatj : estimated latency (time) to open a connection to the server • cbwj : estimated bandwidth of the connection • sclat and scbw : the connection establishment latency and bandwidth for that document are measured • di = clatser(i) + si/cbwser(i) • ser(i) : the server on which document i resides • si : the document's size • di : LAT selects for replacement the document i with the smallest download time estimate

6. Hybrid Algorithm (HYB) • Motivated by Bolot and Joschka’s algorithm W1rtti + W2 si + (W3 + W4 si)/ti • ti : the time since the document was last referenced • rtti: the time it took to retrieve the document • (clatser(i) + WB/cbwser(i))(nrefi** WN)/ si • nrefi: the number of references to document i since it last entered the cache • si : the size in bytes of document i • WB and WN : constants that set the relative importance of the variables cbwser(i) and nrefj

7. Comparison to Bolot and Hoschka • Estimate the quantities on a per-server not a per-document basis. • Estimation algorithm uses history of references, rather than just the last reference. • HYB use the number of accesses to a document, rather than the last access time

8. Workload Mode Replacement Policies Parameter Values 1. VT-CS1 online LAT,LFU,LRU,SIZE N/A 2. VT-CS2,VT-CS3 online HYB,LFU,LRU,SIZE WB=8Kb, WN=0.9Kb,CONN=2Kb 3. VT-CS2,BU,VT-LIB replay HYB,LFU,LRU,SIZE WB=8Kb, WN=0.9Kb,CONN=2Kb 4. VT-CS2,BU,VT-LIB replay HYB See Table 2 -------------------------------------------------------------------------------------------------------------- Table 1 : Experiments performed Trial WB WN CONN 1 8 Kbyte 0.9 2 Kbyte 2 16 Kbyte 0.9 0.5 Kbyte 3 8 Kbyte 1.1 0.5 Kbyte 4 16 Kbyte 1.1 2 Kbyte ------------------------------------------------------------------------ Table 2: Weight settings for Exeriment 4 in Table 1.

9. Variation of download time, workload VT-CS1, Experiment 1.

10. Variation of download time, workload VT-CS2, Experiment 2

11. Experiment 3 HR WHR Time Better Worse Better Worse Better Worse VT-CS2 SIZE* LFU* LRU* SIZE* HYB* LFU* HYB* LRU* LFU* HYB* SIZE LRU BU SIZE * LFU * LRU SIZE * SIZE * LRU * HYB LRU HYB LFU HYB LFU VT-LIB SIZE * LFU* LRU LFU HYB LFU HYB* LRU HYB SIZE* SIZE LRU

12. Experiment 4 • HR is always sensitive to weight WB, • Time is rarely sensitive to the weights, and • WHR is sensitive to WB and sometimes WN.

13. Conclusions • LAT, which estimates download time and replaces documents with shortest download time, • HYB, which combines estimates of the connection time of a server and the network bandwidth that would be achieved to the server with document size and number of document accesses since the document entered the cache. • Suggest : replacement only on estimated download time (algorithm LAT) is a bad idea.