記憶體工作頁面集是一個普遍被運用在作業系統記憶體頁框管理的概念， 大多數行程(process)在其運行的每一個短暫時間區間內僅會使用到數量有限的虛擬記憶體頁面， 而這個記憶體頁面集合並不會隨著時間有很大的變化，這樣的概念主要應用在進行記憶體頁面替換時的捨棄頁面判斷。事實上， 它也可以做為記憶體電源管理的一項重要參考， 譬如待機/休眠時的記憶體保留原則、常閉瞬開記憶體之電源開關管理等。針對記憶體工作頁面集的判別，目前已知的方法皆是在作業系統記憶體管理機制中預先進行記錄，在發生頁面錯誤(page fault)時不斷更新。本論文運用休眠快照的方法將系統凍結來探討某一特定時間點(Android開機)的記憶體工作頁面集數量與分布，這個方法收集當下與未來一段時間內需用到的頁面，且根據我們實驗顯示這個方法具有一致性結果，可以作為記憶體工作頁面集探討的實驗基礎。本論文證實在Android開機完成當下的工作頁面集相對於實際記憶體使用量的比例相當低， 約只有2.63%， 同時歸納出工作頁面集與行程虛擬記憶體區塊之間的初步關係，不同屬性的虛擬記憶體區塊內的工作頁面集與實際記憶體使用量之間並沒有等比例關係，而是有約略相近的數量級關係，大多數行程的工作頁面集數量約在100個頁面以下， 不同虛擬記憶體區塊亦有各自的數量級， 可以做為後續發展記憶體工作頁面集篩選方法的參考。

Working set is a widely used concept for operating system memory management. Most processes use only a finite set of virtual memory pages during each small execute time interval. In addition such a set of
virtual memory pages varies slowly with time. This concept is usually applied to decide which pages to be evicted while executing memory page replacement. However, it can be also applied to be a good reference for
memory power management, such as how to determine the memory pages to be reserved during sleeping or hibernating, and how to manage power on/off on normally-off memory. According to the known algorithm, to find out the working set needs operating system to record the reference status of pages and update records when page fault occurs. This article use hibernation snapshot approach to freeze the whole system to analyze the quantity and distribution of working set pages at a given time (Android boot up). Such a method collects all pages that will be practically used in the near future. Experiments in this article have consistent results so that hibernation snapshot approach can be a well-defined experiment base for studying memory working set.This article has proved that the size of working set of android booting up is much smaller than the size of memory practically used, the ratio is about 2.63%. Moreover, this article also has some basic conclusionsabout the relation between working set and virtual memory areas (VMA’s) of process. The experiment results show that working set in different virtual memory areas have quit different reduced ratios, however, have similar order of magnitude. Most processes have working set of about 100 pages and have different order in different VMA’s. Such conclusion can be a reference to develop the rules for how to find out the working set
pages.