Browse Prior Art Database

Non-Blocking Distributed Bus Switch for Multicomputer Systems

IP.com Disclosure Number: IPCOM000123242D
Original Publication Date: 1998-Jul-01
Included in the Prior Art Database: 2005-Apr-04
Document File: 3 page(s) / 78K

Publishing Venue

IBM

Related People

Bannister, JP: AUTHOR [+4]

Abstract

Disclosed is a design for a bus switch that connects two or more multiprocessor systems. The switch design utilizes dedicated data channels and couples system level channel control with a local transaction reordering mechanism to provide reliable and efficient data transfer.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 52% of the total text.

Non-Blocking Distributed Bus Switch for Multicomputer Systems

   Disclosed is a design for a bus switch that connects two or
more multiprocessor systems.  The switch design utilizes dedicated
data channels and couples system level channel control with a local
transaction reordering mechanism to provide reliable and efficient
data transfer.

   In multi-computer systems comprised of two or more
processing nodes, system performance can be degraded by delays in
the delivery of data to a remote system if a data channel is not
immediately available to the transmitting system.  Such delays may be
the consequence of implementing a data channel between nodes that
serializes data transactions.  A common example of such a channel is
a shared bus.  If multiple devices on a shared bus have one or more
data transactions to transmit, some devices will have to delay
transmission while other devices send data.  Delays may also result
from the data transmission patterns of the system.  If a node has
multiple data packets to deliver to other nodes, and a dedicated
output channel on which to send the packets, these transactions will
be placed on the data channel serially.

   An improvement is to provide dedicated data channels
between node pairs.  This affords immediate access for transmitting
nodes to any and all potential receivers, without consideration for
traffic from other nodes.  This configuration allows simultaneous
transmission of multiple packets, providing efficient multicast and
broadcast.

   A simple...