The specific problem being addressed by Future I/O is the potential bottleneck presented by the I/O bus in the PC architecture. (Why it is thought that I/O makes much real difference in view of the more substantial memory bottleneck isn't addressed in any of the material I've read).
Maximum 1999 PCI bus speeds are approaching 500MB/Sec. A technology called PCI-X promises to double this in the near term future. Beyond that, it is generally felt that I/O will have to be done via switchable point to point connections rather than via a bus. Future I/O (and its competitor-NGIO) are specifications for that point to point connection.
The idea seems to be that various devices will route their signals to the equivalent of a cross-point switch that will connect devices to destinations. Each device will use it's own technology - SCSI, Firewire or whatever and will have an adapter at the switch to achieve FIO compatibility. The switch will apparently be able to route simultaneous non-competing connections between devices that don't require common resources. The initial goal is switching of 1GByte/second connections supporting copper wire connections of up to 10 meters and support for other connection technologies and longer wires via external drivers.
The Future I/O Alliance included 60 companies including Adaptec, IBM, HP, Compaq, and 3Com. However NGIO has a heavyweight list of advocates also.
The current schedule calls for the specification to be promulgated in 1999, with demonstration devices in 2000, and actual products in 2001.
Return To Index Copyright 1994-2002 by Donald Kenney.