Clustering's Path to Platform Independence?

One of the challenges faced by users trying to implement clustering technology is that there are many ways to cluster machines and few standards. Often, you either select a single vendor solution or have to customize the OS and application code yourself.

To overcome this problem, a group of networking vendors has proposed a common way to develop distributed I/O subsystems that will be crucial to the success of clustered systems. The proposal is in the form of a set of specifications (called I²O) that describe an I/O architecture. Typically, you have a host system that requires access to a subsystem (e.g., a disk arra y or a user's terminal display). The traditional way to link the host with the subsystem is to develop customized drivers for the host's OS and drivers for the I/O processor in the system.

The I²O architecture eliminates the dependency on the host OS. Additionally, the architecture is independent of the device being controlled.

Essentially, I²O makes I/O device drivers portable across multiple OSes, processors, and bus technologies. I²O accomplishes this portability by splitting the device drivers in two parts. One resides on the host system and one on the subsystem.

The host portion of the driver is called the Operating System Service Module (OSM). This part of the driver interfaces to host-specific APIs. The subsystem portion of the driver is called the Hardware Device Module (HDM). The HDM is specific to the controller and hardware device that comprise the subsystem.

The OSM and HDM communicate through a piece of software called a communications layer . One of the benefi ts to using I²O is that it simplifies the development of clustered applications. That's because developers do not have to write driver software for every OS and version of an OS.

With efforts like I²O, the industry is finally developing some standards that should make clustering easier to do.