行動商務中快取機制之研究

Abstract

一直以來，較有限的頻寬、較高的延遲、以及較高的傳輸錯誤率都是發展無線網路的首要瓶頸。利用快取記憶體(cache)配合預先擷取(prefetching)動作雖然可以避免無線網路的連結而減少資料傳輸量以提降低延遲，但運用在有線網路的快取機制(cache mechanism)、預測演法(prediction algorithm)卻不適於直接移轉至無線網路中使用。原因則是因為無線網路中無論是快取記憶體(cache)本身的容量或是預先擷取時的資料傳輸量都必須縮小，才不會造成系統的負擔和傳輸上的延遲。 本研究提出一種適用於無線環境的快取機制 — 以使用者行為模式為基礎之路徑預測機制(A User-behavior-based Path Prediction Mechanism, UPPM )。我們利用區分使用者行為模式來切割、組織歷史資料，使歷史資料以及預測範圍縮小，進而使得預測需求的結果也縮小，以達到預先擷取(prefetch)資料量降低的目的。此外，將不同類別的文件資料、歷史資料加以切割，再依不同使用者行為模式對應不同類別之歷史資料的方式來預測使用者需求，也會因為資料的更具代表性、不失真而更提高了快取命中率(hit ratio)。

The limited bandwidth, high latency, and high bit-error rates have been serious bottlenecks in wireless network. Although, caching and prefethcing are the well-known solutions to solve the problems in both wired and wireless network, the cache mechanism and the prediction algorithm in wired network are not fit for wireless environment. The fact we face, wireless network get the smaller capacity of the cache and the higher latency of data transmission compared with wired network, So we have to reduce the volume of the data stored in the cache and prefethched in wireless network. Here we propose a cache mechanism named “A User-behavior-based Path Prediction Mechanism”, which narrows the range of the requested documents predicted by algorithm. The approach analyzes and separates the history data according to different user behavior patterns. Due to the reduced range of history data we refer to at one time, the range of the predicted requests becomes narrow too, and thus the system prefetch a small number of documents. Besides, to classify the history data and to predict requests depending on different user behavior patterns will make history data and predicted results more representative and accurate, and as a result, we improve the cache hit ratio.