本文選題：Memcached　切入點：內(nèi)存替換　出處：《華中科技大學》2016年碩士論文　論文類型：學位論文

【摘要】：隨著數(shù)據(jù)量的增長及應(yīng)用的擴大,基于內(nèi)存的數(shù)據(jù)庫在數(shù)據(jù)訪問過程中扮演越來越重要的角色。內(nèi)存數(shù)據(jù)庫相對于普通持久型數(shù)據(jù)庫有著訪問速度快、輕量、易管理等特點,被廣泛應(yīng)用于各種大型數(shù)據(jù)中心。Memcached作為一種常見的內(nèi)存數(shù)據(jù)庫在Facebook、Twitter、Sina等中得到應(yīng)用,然而Memcached在內(nèi)存替換算法上仍存在無法有效地處理數(shù)據(jù)大小變化較大的不足,當發(fā)生“鈣化”現(xiàn)象時,Memcached會出現(xiàn)命中率降低、訪問延遲增大等結(jié)果,嚴重制約了各種依靠Memcahced作為數(shù)據(jù)緩存的系統(tǒng)的性能。本文對Memcached所處理數(shù)據(jù)的特點進行了研究與分析,發(fā)現(xiàn)不同大小的Item的訪問對一個Slab是否成為一個熱的Slab影響程度不同,而不是以前研究中所提到的固定關(guān)系,同時研究了Slab與Item的關(guān)系,提出了基于影響因子與淘汰代價的Memcached內(nèi)存替換算法。影響因子是根據(jù)不同大小的Item對這個Item所在的Slab是否成為熱的Slab影響程度不同而設(shè)計的,大的Item比小的Item更能使Slab成為一個熱的Slab,進而免于淘汰。同時重新定義了Item平均訪問頻度這一概念:單個item所承受的訪問請求而不是整個Slab;發(fā)現(xiàn)了不同順序的訪問對Slab是否成為熱的Slab影響程度不同這一現(xiàn)象并采用“半衰期”方法量化這種影響,最后利用這兩個因素動態(tài)地更新影響因子的大小。淘汰代價指的是淘汰一個Slab和Item的代價,在Memcached內(nèi)存不足時,通過最近訪問次數(shù)、訪問時間、大小等計算淘汰一個Slab及Item的代價大小來確定是淘汰一個Slab還是Item,而不同于以前研究中提到簡單的主動與被動Slab淘汰策略。改進的方法提高了Memcached在內(nèi)存不足時處理數(shù)據(jù)訪問的能力。測試結(jié)果表明,改進的方法在命中率上有5%左右的提升并將讀寫延遲降低了2%到4.5%。同時改進方案優(yōu)化了Slab和Item淘汰數(shù)量,最后將改進的方法在pNFS中進行了應(yīng)用,測試表明改進的方法降低了pNFS系統(tǒng)的讀寫延遲,最高可達5%。
[Abstract]:With the increase of data volume and the expansion of application, the memory-based database plays a more and more important role in the process of data access. Compared with the common persistent database, the memory database has the characteristics of fast access, light weight, easy management and so on. It is widely used in various large data centers. Memcached, as a common memory database, has been used in Facebook Twitter. However, Memcached still has some shortcomings in memory replacement algorithm, which can not effectively deal with large changes in data size. When "calcification" occurs, the hit rate and access delay will decrease, which seriously restricts the performance of various systems that rely on Memcahced as data cache. The characteristics of the data processed by Memcached are studied and analyzed in this paper. It is found that the access of Item of different sizes has different effects on whether a Slab becomes a hot Slab, rather than the fixed relationship mentioned in previous studies. The relationship between Slab and Item is also studied. A Memcached memory replacement algorithm based on the influence factor and the cost of elimination is proposed. The influence factor is designed according to the influence degree of different size Item on whether the Slab in which the Item is located is a hot Slab. Large Item makes Slab a hotter than a small Item and thus avoids elimination. At the same time, the concept of average access frequency of Item is redefined: the access request to a single item rather than the whole item; the discovery of different order visits. Asking whether there is a different degree of influence on whether Slab becomes a hot Slab and quantifying it using the "half-life" method, Finally, these two factors are used to dynamically update the size of the influencing factors. The elimination cost refers to the cost of eliminating a Slab and a Item. When the Memcached memory is out of memory, it can be accessed through the number of recent visits and the access time. The cost of eliminating an Slab and Item is calculated to determine whether to phase out an Slab or an item, as opposed to the simple active and passive Slab elimination strategies mentioned in previous studies. Ability to manage data access. Test results show that, The improved method improves the hit rate by about 5% and reduces the delay of reading and writing by 2% to 4.5. At the same time, the improved scheme optimizes the number of Slab and Item elimination. Finally, the improved method is applied in pNFS. The test results show that the improved method can reduce the delay of reading and writing of pNFS system, up to 5%.
【學位授予單位】：華中科技大學
【學位級別】：碩士
【學位授予年份】：2016
【分類號】：TP311.13

【參考文獻】

相關(guān)碩士學位論文 前3條

1 楊娣;分布式緩存系統(tǒng)的內(nèi)存利用率和并發(fā)性能優(yōu)化機制[D];華中科技大學;2013年

2 鄧世權(quán);基于Memcached的高速緩存功能擴展研究[D];西南交通大學;2012年

3 倪高鵬;基于Memcached的緩存系統(tǒng)設(shè)計與實現(xiàn)[D];大連理工大學;2012年

，

本文編號：1563275