Browse Prior Art Database

Programmable Error Options for Network Messages

IP.com Disclosure Number: IPCOM000114482D
Original Publication Date: 1994-Dec-01
Included in the Prior Art Database: 2005-Mar-28
Document File: 2 page(s) / 83K

Publishing Venue

IBM

Related People

Olnowich, HT: AUTHOR

Abstract

Disclosed is a method for handling error conditions effectively in switch networks. In a parallel system, wherein a plurality of nodes are interconnected by a multi-stage switch network, each node usually interfaces to the network via a network adapter. The network adapter normally implements message buffers, usually a Send FIFO containing a plurality of messages to be sent to the network, and a Receive (RCV) FIFO containing a plurality of messages which have been received from the network. Error detection mechanisms are usually implemented to detect if errors occur while trying to read messages from either the Send FIFO for transmission to the network or the RCV FIFO for transmission to a local processor.

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

Programmable Error Options for Network Messages

      Disclosed is a method for handling error conditions effectively
in switch networks.  In a parallel system, wherein a plurality of
nodes are interconnected by a multi-stage switch network, each node
usually interfaces to the network via a network adapter.  The network
adapter normally implements message buffers, usually a Send FIFO
containing a plurality of messages to be sent to the network, and a
Receive (RCV) FIFO containing a plurality of messages which have been
received from the network.  Error detection mechanisms are usually
implemented to detect if errors occur while trying to read messages
from either the Send FIFO for transmission to the network or the RCV
FIFO for transmission to a local processor.  A problem occurs when
these message errors are detected as to how to handle them
efficiently with respect to the message protocol being used.  There
are numerous message passing protocols which fall basically into two
error handling categories: 1) The software is responsible for
tracking, reporting, and trying to correct errors, or 2) The data
flow is the important parameter and the software is to immediately
throw away the faulty data and move on.

      This solution is to provide hardware features in the network
adapter to aid the software in providing an efficient error handling
scheme.  In specific, the adapter hardware provides two programmably
selectable options associated individually with each FIFO: Halt on
Error or Throw Away.  Thus, the adapter is flexible in implementing
both options, and the software is given the selection on an
individual FIFO basis.  For network adapters which use multiple Send
and RCV FIFOs, the software can have the option of selecting the
"Halt on Error" option to apply to one FIFO set, while selecting a
second FIFO set to implement the "Throw Away" policy.  Thus, the
software can control different types of message protocols in the same
network adapter, and even do it simultaneously.

      Halt on Error:  The "Halt on Error" philosophy being
implemented at the network adapter causes the adapter to stop
processing messages in a FIFO that has experienced an error.  The
hardware disables the FIFO from further attempts to be read or
processed, leaves the failed m...