OSD: The Future of Storage

Information Management Journal, Jan/Feb 2005 by Straub, Joe, Samarra, Kenneth

Object-based storage device (OSD) technology promises to change forever the enterprise architecture world by making upgrades easier and more affordable

Enterprise architectures is the hot new term in information technology (IT) circles. It refers to a comprehensive system overview that includes the entire IT structure for an organization - including mainframes and permanent data storage, along with all support software, and the servers and work stations connected via WAN/LAN (wide area network/local area network).

It is appropriate that the entire enterprise is now getting attention because, in the past, ad-hoc improvements to specific pieces of computer system equipment or upgrades to system software often had detrimental effects on the overall operation. In fact, most IT professionals have horror stories to tell about upgrades costing more in time, frayed nerves, and value than if the original system had remained in place without modification.

By focusing on the entire enterprise and adjusting the components with consideration for system bottlenecks, a few well-chosen decisions can often significantly reduce the time required to install or upgrade software versions. The enterprise architecture approach also prevents unintended consequences associated with narrowly focused computer system modifications.

Object-based Storage Devices

One of the most important changes in years is now taking place within the enterprise architecture community in the form of object-based storage device (OSD) technology. OSD will likely make enterprise upgrades much easier and more affordable for a variety of users.

Just as object-based programming forever changed the software development world more than a decade ago, OSD promises similar achievements for enterprise architectures.

OSD provides a straightforward approach to connecting all enterprise data storage peripherals. OSD simply assigns more control of data movement directly to the storage devices. This means that user requests are handled at a high level (as objects) and each OSD peripheral device can respond as needed to optimize local performance. For example, in the client-server world, a request for data goes from the client workstation to the server, which then passes it to a storage device, retrieves the needed data set, and passes it down to the client workstation for viewing and use. In the OSD scenario, the request goes from the client workstation directly to the storage device, which has enough intelligence to do the work previously performed by the separate server.

The implication is this: In client-server technology, each client workstation must have software installed on it that is compatible with the server. When software is upgraded, each server and each client workstation must be upgraded, too. With OSD, the upgrade occurs at the storage device.

In addition, there are currently myriad proprietary technologies to manage storage. With OSD, each of these individual technologies or operating systems can coexist within the same network.

OSD has been made possible by the availability of powerful computer capability in small chips. Much of the intelligence that was contained in a central source (the mainframe) in the past can now be shifted to each peripheral device. Although this capability has been available for many years, computer system architectures are slower to adopt it because they are built on a variety of standards that cannot be easily changed.

IT researchers at Carnegie Mellon University's Parallel Data Lab (www.pdl.cmu.edu), under the direction of Garth Gibson, have been working on this new technology that radically overhauls the structure of storage as we know it. Many have called OSD the "Holy Grail of Storage," mainly because it allows for self-managed storage devices.

Past Imperfect

Over the past 30 years, the basic components of data storage architectures have largely remained the same. There is the software application that users "order" to access certain data for a computational routine. Applications have input/output (I/O) requests of varying complexity. Requests are submitted to a file storage management system (FSMS), as shown in figure 1, that organizes and prioritizes all requests. The FSMS's file system storage component then directs the movement of data to and from physical storage devices. Finally, there is the storage device itself (typically disk or tape drives) where the data physically resides on some media for action now or in the future.

In the mainframe era of the past, storage devices were attached directly to the computers they served via a method known as direct attached storage (DAS). DAS featured various storage layers but there was no network- the storage device is attached to the host computer. The one-to-one relationship between host computer and storage device made this type of attachment simple to install and manage, with the file system storage component talking directly to the storage device (see Figure 2). Both tape and disk could be configured in this way. But because of the fundamental limitations of the architecture, DAS devices could not talk to more than one computer at a time. DAS used block-based storage, a method in which files are chopped up into chunks of data that are written as blocks.