自動協調型分散式記憶體快取系統

Abstract

當網路服務日趨複雜，資料庫系統也愈趨龐大；此時單憑資料庫系統已不足以應付日漸增加的使用者需求。此類問題有許多解法，其中使用記憶體快取系統是一個常見的方案。記憶體快取是一種藉由將資料與物件儲存在記憶體內，藉此減少存取資料庫次數與硬碟讀寫時間而達到加速效果的技術，並被大幅應用在大型網站系統。Memcached 則是在記憶體快取系統中最常使用的一種服務。在本論文中，我們設計了AOM：一套基於Memcached的具延展性之分散式記憶體快取系統，以符合日益進步的雲端環境。AOM具有無縫接軌的特性，可在不更動現行系統架構或程式碼的情形下對Memcached伺服器集群進行擴充或縮減。此外，AOM更會自動分配所儲存之物件，並一定程度達成伺服器群集的負載平衡。AOM除對Memcached實現具延展性的擴充能力外，其本身架構也具備可調配性，可依不同情境對AOM架構進行調整。AOM更不需增加使用者開發成本，甚至於特定情形下可減少使用者開發流程，協助使用者進行自動化的物件散佈。實驗結果顯示，雖然在增加AOM之架構下須付出一定系統成本做為交換，但考慮其便利性、擴充性與自動協調性，我們認為在一般情境下其作為交換的效能成本是可被接受的。簡言之，AOM可在不更動網路服務現行架構的前提下，完成功能性的擴充並降低開發成本。

When network services become more and more complex, the size of database systems also grows larger and larger. Database systems thus cannot satisfy the need of the users. Memory cache system is a common solution to this modern problem. It keeps the data and the objects in the memory to reduce the requests send to the database system. Since the speed of memory is faster than disk I/O, memory cache system effectively reduces the query time needed. Memcached is the most common package used in memory cache systems. In this thesis, we introduce Agent Of Memcached, AOM. It is a distributed, standalone daemon that helps Memcached to achieve scalability and conveniences. AOM can seamlessly integrate into current service architecture without modifying client code or server code. It can also add new memcached server or remove dead memcached server from the online service without stopping the whole service, and achieve scalability for Memcached. AOM will automatically distribute cached objects, thus reduce the workload of client-side and save development time. Although there are performance tradeoffs to add AOM into online services, we believe the scalability, convenience and co-operating ability that AOM brings is worth the cost. In short, AOM can achieve functionality expansion and lower development cost without changing current architecture.