Research On Key Technologies Of Object-Based Active Storage

With the development of computer and Internet, new trends have emerged in network storage: the amount of data increases more quickly, storage management needs more intelligence, and cross-platform data share becomes more important. Object-Based Storage(OBS) is an emerging standard designed to address this problem. Using object-based interface, OBS can overcome the limitation of block-based and file-based interface, and can improve safety, data sharing, scalability and performance. The access unit in OBS is object. An object contains not only user data, but also attributes describing characteristic of data. Object-based interface is more expressive than others, because it can transport attributes among clients and devices.At the same time, devices have embedded with more and more processing capability, more funcionts are migrated to devices(active storage). However, these devices are usually"dump", which response to outside requests passively. In this way, devices are externally managed without involving in storage management. To make management more efficient, devices should become more intelligent, self-managed and application-aware. In fact, this problem can be solved by exploiting active storage technology. Moreover, disks are still slow devices; and delivering data throught WAN still brings non-neglective delay. To avoid this, data should be processed inside storage devices using downloaded code. With the emergency of object-based interface, active storage can take advantage of expressive interface to achieve the cooperation of application and devices. Object-Based Active Storage System (OBASS) is the combination of OBS and active storage, and will be the next wave in network storage field.OBASS is built on the basis of objects, and all the information will be represented as objects. Objects are storaged in Object-Based Storage Devices (OSDs), which is the building block of OBASS, The software in OSD mainly composes of three layers: object layer, active service layer and QoS layer. Object layer is responsible for object data organization in disks. Active service layer implements active storage, and application can download code to devices saveing as special objects (method object). Active service layer is responsible for scheduling of methods and puts methods into execution. The QoS layer controls all software layers to satisfy the QoS demands of all kinds of users under various workload.In object layer, data organization in one disk is implemented in Long-lived Object-Based File System(LOBFS). One goal of LOBFS is to achieve persistent performance, that is, the performance doesn't decline after a long period during which a large number of operations (object create/delete/write/read) are performed in LOBFS. The persistent performance is implemented through flexible free space management and gradual space allocation which uses variable allocation granulality.In object layer, another data organization problem is object assignment among disks (object placement). Different application adopts different placement policies. Two policies, based on blocking probability model and response time model, are presented to use in stream media and transaction processing application respectively. In OBAS, object placement policy can be used in on-line circumstance. To address the online problem without a priori knowledge of workload parameters, the policy employs an adaptive mechanism to estimate the characteristics of workloads based on device intelligence and object attributes.In active service layer, T10 OSD protocol is modified to support active storage, including defining new object such as method object. Method object can be put into execution in two approaches: request-driven and condition-driven. Combining these two approaches, the active service layer establishes a uniform framework for offloading computing task and management task to OSDs. Method execution is the key of active storage, an implementation in Linux is illustrated for two kinds of method: filter-typed method and service-typed method. To evaluate the effect of active storage, two applications are investigated: OSD File System and load balancing. In the first application, two operations(lookup and unlink) are offloaded to OSD as filter-typed method, and the analysis shows that network delay and memery copy can be reduced largely. In the second application, load balancing algorithm can be executed as service-typed method in OSD, experiments show that the system response time can be sharply reduced with object replication and migration.Storage QoS control can be achievd through three layers: Upper layer(object level) implements the scheduling of object request; Middle layer(object-block level) implements QoS through object prefetching, page preallocation, delay space allocation and delay writing; Lower layer(block level) schedules block requests for both realtime workload and non-realtime workload by considering three factors: deadline, disk head position and I/O priority. The tests show that the jitters are at least 1(for realtime read) or 2(for realtime write) order of magnitude less than the ones under other schedulers.