What is an offloaded data transfer? How does it differ from regular file transfers, and how can an enterprise benefit from it?
Storage is a critical resource in any data center environment, but storage is also the slowest resource. It often imposes limits on workload behavior and performance, data protection processes and other Windows Server activities. Technologies that accelerate storage, such as caching and tiering, can yield significant performance benefits. But storage still demands the direct intervention of servers to read, write and move data between disks or storage arrays. Emerging storage systems can speed data transfers while easing the computing overhead on servers. Consider how a normal file transfer works. The originating server must first read data from storage, transfer that data through the LAN to a destination server, and then the destination server writes data to a different storage location. This process introduces a great deal of latency and requires computing resources on the source and destination server. It's inefficient -- especially when data movement occurs regularly across the data center.
Windows Server 2012 introduced the notion of offloaded data transfer (ODX), which can move data within the same storage array or move data between different storage arrays without the need to shuttle the data to or from servers. This lowers the computing overhead needed for data transfers and improves apparent performance during the copy/migration processes. These benefits are particularly noticeable when transferring large files (such as streaming media) or loading/saving performance-sensitive files like virtual machine disk files.
An offloaded data transfer relies on the use of tokens which represent file data. For example, when an ODX file transfer takes place, Windows Server 2012 converts the transfer into an ODX request and the originating server receives a small token. The small token is then moved to the destination server and on to the destination storage system. After receiving the ODX token, the destination storage system coordinates the move directly from the originating storage system. Once a token is passed, the data transfer takes place between storage systems (or locations within the same system) without any intervention from either the originating or destination server.
This was first published in March 2014