Skip to Main Content
The concept of active storage was investigated through adding computational capabilities to I/O contexts, in which increasing data-locality in parallel or placing post-/preprocessing functionality for storage self-maintainability. The advantages of this scheme are obvious, because it is able to leverage the processing ability of the storage nodes and considerably reduce the network traffic at the same time. The previous research efforts on active storage are either based on file or block interface. Neither of them is able to be a suitable balance between content understanding and data locality for parallel processing. The emergence of object interface provides a perfect solution for the development of active storage, and the functionality and scalability of active storage in the object storage system, The active storage approach that we proposed is targeting object-based storage system upon the building block of extended international standard T10 OSD (Object-based Storage Device) command set. Different with the existing work, our work focuses on researching and developing two new kinds of method object to realize active storage, in which user method object is used to manipulate the common user data object, dynamically adjusting I/O performance by the user's read and write requests, and the system method object is used in accordance with the policy object and status of the system to implement self-management of the system respectively. Then these active objects are put together to integrated into proposed active storage framework. Experimental results show that the method of active method objects and runtime framework, which is to develop policies and mechanism for active storage, can reduce the amount of network data transmission and user response time, greatly improving performance and intelligence of the system.