【摘要】：信息技術(shù)正在從以計(jì)算設(shè)備為核心的計(jì)算時(shí)代進(jìn)入到以數(shù)據(jù)為核心的存儲(chǔ)時(shí)代.數(shù)據(jù)量和數(shù)據(jù)價(jià)值的增加,對(duì)存儲(chǔ)系統(tǒng)的容量、成本、訪問(wèn)性能和數(shù)據(jù)可靠性等方面提出了更高的要求.為減小冗余數(shù)據(jù)開(kāi)銷(xiāo)并保證數(shù)據(jù)的可用性和可靠性,采用糾刪碼技術(shù)已成為業(yè)界的發(fā)展趨勢(shì).數(shù)據(jù)一致性是糾刪碼存儲(chǔ)系統(tǒng)的本質(zhì)屬性,保證了數(shù)據(jù)的可靠性和可用性,高效保證糾刪碼數(shù)據(jù)的一致性是糾刪碼系統(tǒng)的關(guān)鍵.提出"寫(xiě)后重映射"存儲(chǔ)模型(After Write Remapping,AWR)和事務(wù)實(shí)現(xiàn)方法,保證了糾刪碼數(shù)據(jù)的一致性.AWR模型為糾刪碼數(shù)據(jù)的更新過(guò)程提供原子性和持久性,采用非原位更新和原子重映射機(jī)制保證了更新過(guò)程中數(shù)據(jù)的一致性,具有通用性好、存儲(chǔ)開(kāi)銷(xiāo)低和事務(wù)性能高的特點(diǎn).經(jīng)過(guò)理論分析和實(shí)際測(cè)試驗(yàn)證,AWR事務(wù)存儲(chǔ)模型帶來(lái)的糾刪碼存儲(chǔ)系統(tǒng)讀性能損失低于1%,64KB及以上粒度的寫(xiě)性能損失低于3%,高效地保證了糾刪碼數(shù)據(jù)的一致性.
[Abstract]:Information technology is moving from the computing era with computing equipment as the core to the storage era with data as the core. With the increase of data volume and data value, the capacity, cost, access performance and data reliability of the storage system are required higher. In order to reduce the cost of redundant data and ensure the availability and reliability of data, erasure code technology has become the trend of the industry. Data consistency is the essential attribute of erasure code storage system, which ensures the reliability and availability of data. The key of erasure code system is to ensure the consistency of erasure code data efficiently. The storage model (After Write Remapping,AWR) and the transaction implementation method are proposed to ensure the consistency of erasure code data. Awr model provides atomicity and persistence for updating erasure code data. Non-in-situ updating and atomic remapping mechanisms are used to ensure the consistency of the data during the updating process, which has the characteristics of good generality, low storage overhead and high transaction performance. Through theoretical analysis and practical test, it is proved that the performance loss of erasure code storage system caused by awr transaction storage model is lower than 1 / 64KB and above granularity and the write performance loss is less than 3KB, which effectively ensures the consistency of erasure code data.
【作者單位】： 中國(guó)科學(xué)院計(jì)算技術(shù)研究所;中國(guó)科學(xué)院大學(xué);
【基金】：中國(guó)科學(xué)院先導(dǎo)專(zhuān)項(xiàng)基金項(xiàng)目(XDA06010401)資助 天津市科技計(jì)劃項(xiàng)目(15ZXDSGX00020)資助
【分類(lèi)號(hào)】：TP333
，

本文編號(hào)：2250910