| With the emerging of data intensive supercomputing era, contemporary research on computer system has shifted the focus significantly from the computing-centric model to the data-centric model. This trend imposes a increasing demand on the reliability, manageability and performance of storage systems. Aiming to satisfy different application needs, the requirements for storage systems also vary dramatically. Since Object Storage Device (OSD) based storage systems are capable of optimizing the data organization according to various characteristics of data accesses patterns originated from different types of storage applications, taking advantage of the intrinsic autonomicity, agility and intelligence of each individual storage devices to improve the quality of storage services (QoSS) at the device level, nowadays they inevitably becomes one of the most promising storage system solutions targeting various storage application needs at one time.By analyzing multi-users concurrent access patterns and the performance of multimedia applications, a high-performance architecture of data organization for multimedia object storage device (MOSD) is proposed. MOSD adopts a two-level architecture: the lower layer organize the data in a RAID-5 style in order for high disk space utilization; the upper layer stores the "hot" object. By placing different types of objects on different layers, the various type of user requests are effectively distributed to discrete corresponding disks. In simulation tests of multi-user environment, the performance of MOSD is higher from 90% to 550% than that of RAID0.Based on the performance analysis of Web storage applications, this paper presents a data organization architecture for Web Object Storage Device (WOSD). The hierarchical storage system designs aim to divide the storage space into two layers: the lower layer, referred to as Cache disk, and the upper layer as the normal storage space. Additionally,this paper devises a Web objects' correlative algorithm, when system is idle, trace information are analyzed,the algorithm uses a sliding window method to calculate the temperature of the correlation among objects in the traces (The temperature values of correlations are only for the objects are in the sliding window),then creats the different correlation sets according to correlaive coefficients, and moves different correlation sets to Cache disk, it minimizes not only the distance of disk head movements in the array, but also the frequent seek operations,hence improves I/O performance of array. The test results show that WOSD delivers great performance - it improves the performance in terms of average I/O throughput by 20% to 25% in multi-users concurrent access environment.In addition, based on a thorough analysis on the fault-tolerance capability on a variety of existing storage systems, a architecture of data organization for high availability object storage device (MOSD) is proposed. The HAOSD is composed of two layers: The upper-layer achieves the high availability by storing multiple replicas for each storage object in a set of storage devices. The individual replicas can service the I/O requests in parallel so as to obtain high performance. But the effective disk space utilization rate for the upper-layer is relatively low. The lower-layer deploys RAID5, RAID6 and RAID_Blaum coding schemes to tolerate multi-disk failures. The disk utilization rate of coding schemes is higher than that of multiple replicas. The HAOSD puts new objects and hot objects in its upper-layer, so that the majority of the requests are absorbed by the upper-layer, hence achieving guaranteed system I/O performance. The main purpose of the lower-layer is to provide a reservoir for the cold data. The simulation tests shows that HAOSD delivers great performance when ensures reliability, In a 1000Mb Ethernet interconnection environment, the maxium read and write performance for a HAOSD server reaches 104MB/s.
|
PDF Full Text Request