有鑑於快取空間的排擠效應以及各網頁物件群組的再用程度不一等因素下，本研究提出一「快取排它策略」應用於網頁代理快取伺服器(WebPCS)，藉以提昇快取系統效能。 本研究主要的方法如后，1）先透過Proxy 2日誌檔為分析樣本，來分析WebPCS之群組式快取效能。2）依前項數據，再確立出「快取排它門檻」條件，用以篩選網頁物件作為快取對象。本快取效能模擬之比較對象為傳統LRU vs. 排它式LRU。 經由模擬結果得知，當排它門檻條件設為C₃、C₄或C₅之時，有高於99.5％用戶端可獲益於WebPCS之命中率□上昇與平均等待時間□下降之優點；但是，最多只犧牲27％用戶端可能受損於命中高容量物件的機會而導致WebPCS之資料量命中率□下降。此一研究成果證實，將排它策略置於快取替換機制之前，可預期改善系統的快取效益。

This paper proposed a novel "cache eviction policy" (CEP) designed for web proxy cache server (WebPCS) to improve the performance on caching objects. First, we investigated the effects of grouped objects on cache performance of WebPCS through the on-line access logs of Proxy2 supplied by NTUST. Thus from the data investigate das above, we could set the values of threshold with which the evictor could determine whether to evict objects the server downloaded. Last, we have selected two types of replacements policy, the conventional LRU (LRU) an evicted LRU (LRU-evict), to be evaluated by the simulation system for comparing their cache performance. The experimental results shows that more than 99.5% of clients may obtain more benefits, including the hit ratio (□) rose and the average latency time (□) fell, as the threshold conditions of LRU-evict were set to be C₃, C₄, or C₅. On the other hand, at most only 27% of clients may lose a chance to hit high-volume objects; therefore, the byte-hit ratio (□) would become lower. These findings confirmed that the overall performance of cache system could be promoted by locating the CEP ahead of the replacement mechanism like LRU for intercepting low reusable objects.