PC blades move the PC hardware into the data center, leaving a thin client on the desktop that essentially functions as an extended keyboard, monitor and mouse. Like server blades, PC blades fit into a chassis that can be centrally managed. But just putting users' PCs in the data center misses the point. "I don't think there needs to be a one-to-one relationship between users and PCs," the CIO at a large insurance company once told me. "Isn't there an enormous amount of processing power that's wasted in that?"
Blade vendors finally have an answer.
Like server blades before them, PC blades are becoming consolidation platforms. For example, ClearCube Technology in Austin offers a management tool for its PC blades called Grid Center that allows up to five users to share a single blade. It also maps users to a pool of PC blade resources that can be dynamically allocated to any user rather than being dedicated to a single user's thin terminal. Meanwhile, Hewlett-Packard is working on technology in its Consolidated Client Infrastructure that will enable administrators to create different tiers of PC blades and provision them based on users' roles.
IBM has taken PC blade virtualization even further. Its Virtual Hosted Client Infrastructure brings the PC blade and server blade functions under a single architectural umbrella. The system enables a thin client to access a PC virtual machine running on a server blade within a BladeCenter chassis. The design uses EMC's VMware and ClearCube's Grid Center to support up to 14 PC sessions per blade. Such consolidation offers much greater efficiencies than just simplifying PC management by centralizing the system hardware. The rigid, "one user to one PC blade" design is disappearing.
It gets better. Today, users attached to a failed ClearCube blade can be quickly moved over to another available blade. The process requires administrator intervention, but ClearCube says it has nearly completed work on an automated fail-over process that avoids interrupting the user session. The next logical step might be to provide tools that let administrators "lift and shift" live user sessions between PC blades for maintenance reasons -- something administrators can do with server blades today using VMware's VMotion.
IBM sees room for further consolidation. Today, a user can be mapped only to the virtual machines that reside on a single blade. By the third quarter, users will be able to connect to virtual machines running on any blade in a chassis. Beyond that, IBM envisions pooling resources in a "gridlike manner," says Juhi Jotwani, director of solutions and alliances for the xSeries and BladeCenter. "Users will still have the personalization, but resources will be spread across the entire chassis," he says.
Ultimately, the PC session itself may be broken apart and processed in parallel on one or more blades. This will evolve naturally as computers move first toward dual-core and then to multi-core processors, and as software developers begin to optimize for the new processor architectures by allowing more and more operations to execute in parallel.
Windows sessions that can leverage a PC blade grid may be a ways off, but opportunities to leverage PC blades for grid computing aren't. Some IT organizations are already using desktops after hours for grid computing tasks. Not only will PC blades in neatly aligned racks in a data center be much easier to set up and manage as a grid, but the design also presents opportunities to speed I/O between PC blades to improve performance.
While using BladeCenter server blades today might sound like overkill for PCs, that's likely to change if blade designs adapt to general-purpose use. As servers and PCs are abstracted from underlying hardware through virtualization, the distinction between what constitutes a PC blade and a server blade will fade. "Blade PCs are going to blur the definition of servers," says Tad Bodeman, director of blade PC and thin-client solutions at HP.
Ultimately, blade PCs will simply appear as instances on one or more virtual machines that get assigned to segments of a general compute-resource pool. The thing the user thinks of today as the PC will simply be absorbed into it.
-- Robert L. Mitchell is a Computerworld national correspondent. Contact him at robert_mitchell@computerworld.com.