基于QoS的數(shù)據(jù)遷移模型的設(shè)計

結(jié)合對象存儲的特點，提出基于QoS 的存儲系統(tǒng)模型。該模型將遷移任務(wù)劃分為細粒度的遷移請求，使對象存儲設(shè)備在實現(xiàn)數(shù)據(jù)
遷移的同時能響應(yīng)I/O 請求。元數(shù)據(jù)服務(wù)器按相同的標準給I/O 請求和遷移請求分配相應(yīng)的收益，使對象存儲設(shè)備能采用收益最大算法調(diào)
度I/O 請求和遷移請求，從而提供更高的服務(wù)質(zhì)量。通過收益預(yù)測和帶寬預(yù)留實現(xiàn)在線最大收益調(diào)度算法。實驗表明，最大收益算法與通
常的遷移優(yōu)先算法和固定平均遷移率算法相比，對系統(tǒng)的I/O 性能影響最小。
關(guān)鍵詞：對象存儲；服務(wù)質(zhì)量；數(shù)據(jù)遷移；最大收益
Design on Data Migration Model Based on QoS
LUO Yi-hui1,2, XIE Chang-sheng2
(1. School of Mathematics and Computer Science, Hubei University, Wuhan 430062;
2. National Storage System Laboratory, School of Computer, Huazhong University of Science and Technology, Wuhan 430074)
【Abstract】A storage system model is given based on Quality of Service(QoS) according to the object storage characteristics. This model divides
the migration task into some migration requests, which makes object storage device respond I/O request as it migrates data from one site to others. At
the same time, the metadata server assigns the income to the I/O and the migration request according to the same criterion, which make the storage
devices respond the both request by the maximum income algorithm and provide the better QoS. The maximum income algorithm is online realized
by income forecasting and bandwidth obligating. Experiment proves that the maximum income algorithm has less infection to I/O performance than
the migration first algorithm and the fixed migration rate algorithm.
【Key words】object storage; Quality of Service(QoS); data migration; maximum income