High-speed networks create pressure on servers for improved system level performance. The system I/O performance must keep up with the high-speed network protocols, such as 100-Mbit Fast Ethernet connections. Distributing the I/O processing between the host processor and an intelligent I/O processor balances the server processing for optimal performance. This paper shows system architects some of the issues associated with distributing I/O functions between the host processor and an intelligent network I/O subsystem. Distributed processing enables the system architect to off-load interrupt processing, device specific functions and data algorithms to the intelligent I/O subsystem. This offloads elements that dont scale with the host processor frequency, such as host processor reads from the PCI bus. Some of the issues associated with distributed processing that are addressed include device driver communication, data ownership/movement, and system configuration.
152356 bytes
pci96.pdf
Legal Stuff © 1997 Intel Corporation
