Traditionally, server performance has scaled to meet demand with the addition of higher-speed processors, symmetric multiprocessing and system improvements in system architecture such as faster buses and memory. Because it makes demands on the server's host processing resources, heavy I/O traffic is not only creating a serious performance bottleneck, it also limits the scalability that can be achieved by even the most innovative host processor and system improvements.

The I₂O architecture specification defined by the I₂O SIG* (Special Interest Group) provides a solution by allowing I/O to function independently from both the specific device being controlled and the host network operating system (OS). I₂O technology allows a combination of device driver standardization, increased I/O throughput and increased scalability, while establishing a foundation for innovations in future server performance.

Device driver standardization
Multiple OS alternatives are now in widespread use, including NetWare 4*, Windows NT* Server and UnixWare*. With traditional I/O architecture, a new driver must be developed for every unique OS and device combination.

I₂O technology comes to the rescue with a "split driver" model that makes device drivers portable across different OS and hosts combinations. The I₂O driver stack consists of an Operating System Services Module (OSM) and a hardware device module (HDM). Once the OS suppliers write a single I₂O technology-ready driver for each device class (i.e. storage, LAN/WAN, etc.), device manufacturers need only to write a single I₂O technology-ready diver for their device, which is then portable across any I₂O technology-compliant OS.

Increasing server capacity
I₂O technology delegates I/O requests to multiple "smart" subsystems for administration by independent I/O processors (IOPs), without intervention by the host processor. Intel i960® I/O processors integrate a RISC processor core controller, a PCI-to-PCI bridge and support for interrupt management and DMA transfers. Distributed I/O processing improves throughput by relieving the host processor of many mundane interrupt-intensive I/O tasks. At the same time, it makes additional host processor resources available for high bandwidth client/server applications.

Because tasks are offloaded from the CPU, I₂O technology can enable the server to handle a larger application load, or more users. This architecture also removes the bottleneck to scalability in server design. Adding more symmetric multiprocessing capability and more smart devices to the server can result in a linear improvement in performance.

Building a foundation for I/O innovation
With its split driver model, I₂O architecture is a spur to the development of a new range of cross-platform intelligent I/O devices. By separating the server application from I/O processing, I₂O technology sets the stage for dramatic performance-enhancing innovations, including IA-64 processors, in addition to improvements in system and memory architecture.

The I₂O technology stackable driver model allows developers to insert specialized functions in an Intermediate Services Module (ISM) which resides between the OSM and the HDM. Examples include compression/decompression algorithms, RAID controller functions, application and system management, data encryption, and support for Internet firewalls.