Distributed vs. Centralized Storage: The Struggle Continues

IBM recently announced iBoot, a new storage technology that enters the ongoing struggle between distributed and centralized storage strategies. In essence, iBoot, which IBM's Haifa Research Laboratory developed in Haifa, Israel, brings to desktop computing many of the same concepts that have long informed mainframe computing. Instead of storing information locally on a hard disk, individual users designate storage space on a larger disk on the network. The disk appears to be local, but it isn't.

Fundamentally, the iBoot technology lets desktop computers operate without a local hard disk. Instead, the computer uses the Internet SCSI (iSCSI) to connect to a hard disk over the network and can boot up Windows XP, Windows 2000, or the Linux OS by using a remote machine. IBM researchers say that iBoot is the first technology to support the remote booting of any Windows OS. Although technologies such as Etherboot, an open source software (OSS) application, and Intel's Preboot Execution Environment (PXE) have supported remote boot for Linux, IBM researchers claim those approaches aren't transparent to the end user. Moreover, Etherboot and PXE require that the entire booted OS be sent to the diskless PC and retained in memory. Users can't make any permanent changes to the OS kernel.

With Etherboot and PXE, a client computer obtains the location of a target computer, then, by using the Trivial FTP (TFTP), downloads a root file system and OS kernel and executes the OS kernel. IBM iBoot uses a different approach. For iBoot, however, IBM researchers have created a small-option ROM image that contains iSCSI client code, a TCP/IP stack, and BIOS interrupt code. When a user turns the machine on, the BIOS disk I/O interrupt uses the iBoot code to communicate directly with a remote drive. With this approach, users can make and retain permanent changes to the OS kernel. Although storage can then take place centrally, each desktop still retains its customized computing functionality.

You can easily envision the advantages of consolidated desktop storage. Hard disk crashes have been a nightmare for a generation of users and systems administrators. One centralized storage space for desktop computing would make backup and restore operations much easier to manage.

Desktop storage is an underutilized resource in many enterprises. As hard disk capacities continue to grow, each desktop contains more and more unused space. A standard desktop computer configuration in many enterprises now includes a hard disk with a minimum of 20GB. Disk drives with 80GB to 120GB hard disks are common. Many users use less than 25 percent of that capacity, and they'd use even less if they had an incentive to erase data they no longer need. However, as rich media use becomes more commonplace, users will need more storage to handle their audio and video files. Centralized hard disks would make it much easier for storage administrators to control that growth. Moreover, desktop storage centralization can reduce the cost of storage upgrades—every desktop would no longer need an upgrade. In the long run, iBoot could impact the design of many classes of computers, from servers to handhelds. IBM researchers have suggested that in the future, computers might come with very small hard disks designed to hold temporary files or small amounts of information (or in the case of servers, no disks at all).

Currently, iBoot is a technology, not a product, and iSCSI network cards are only now being introduced into the market. Although most pundits think that iSCSI will soon catch on as a significant network storage protocol, its rate of acceptance is unknown. Nevertheless, iBoot is another technology that will require storage administrators to clearly think through organizational and end-user storage needs. Increased storage capacity isn't the issue; whether that capacity is centralized or distributed is.