Browse Prior Art Database

Message Passing Protocol for Personal Computer Systems

IP.com Disclosure Number: IPCOM000106563D
Original Publication Date: 1993-Nov-01
Included in the Prior Art Database: 2005-Mar-21
Document File: 6 page(s) / 224K

Publishing Venue

IBM

Related People

Milenkovic, M: AUTHOR

Abstract

Described is an architectural implementation to provide a message passing protocol for personal computer (PC) systems equipped with a Micro Channel* (MC). The protocol allows the efficient delivery of bulk data through a message passing transport facility by shuttling data blocks between participating peers.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 31% of the total text.

Message Passing Protocol for Personal Computer Systems

      Described is an architectural implementation to provide a
message passing protocol for personal computer (PC) systems equipped
with a Micro Channel* (MC).  The protocol allows the efficient
delivery of bulk data through a message passing transport facility by
shuttling data blocks between participating peers.

      Typically, the use of message passing facilities are confined
to high end computer bus implementations and not generally in the PC
class of environments.  The concept described herein implements a
message passing protocol for PC systems equipped with the MC such
that the protocol will enhance MC utilization without changing the MC
architecture.  This is done through the message passing protocol
which is designed to increase the efficiency in the transfer of data.
It is useful in applications such as; multimedia, isochronous data
delivery, and high-end desktop and server applications where
significant performance benefits may be achieved by using deep
adapters with the capability of handling efficient peer-to-peer bulk
data transfers.

      Generally, subsystem control block (SCB) architecture is used
to govern peer-to-peer operation on the MC.  The SCB is considered to
have low protocol overhead and the use of the MC block transfer
cycles amortizes the overhead of bus arbitration over all bytes in
the block.  As a result, message passing has high data transfer
efficiency, typically on the order of 90%, whereas other modes of
data transfer, such as direct memory access (DMA), have lower
transfer efficiencies.

      Message passing is performed by using the MC block transfer
cycles, such as streaming or bursting, to move data between peers.
Address spaces of the sender and the receiver are fully decoupled to
facilitate extension of message passing to clustered and distributed
environments.  The receiver appears in the address space of a sender
as a single input/output (I/O) address port to which all data blocks
intended for the related receiver are addressed.  In a typical
implementation, the receiver contains a hardware first-in-first-out
(FIFO) connected to its message port.  All incoming blocks addressed
to the receiver's message port are staged in the FIFO for subsequent
dispatching to the receiver's memory.  In the case of a deep adapter,
it is connected to an I/O device, such as a network or a direct
access storage device (DASD).

      There are two basic types of messages:  Control messages and
data messages.  Control messages are generally short and data
messages can be long, up to 16 GB and are fragmented into smaller
chunks of data fragments so that each chunk can be transferred across
the MC within a single arbitration window.  This is typically the
duration of MC ownership with no preemptions guaranteed to a bus
master following an arbitration win.  Fragmentation of data at the
sending end and the assembling at the receiving...