Browse Prior Art Database

Time Clustered Access for Maximizing Bandwidth Utilization via Widows of Asynchronous Arrival

IP.com Disclosure Number: IPCOM000113347D
Original Publication Date: 1994-Aug-01
Included in the Prior Art Database: 2005-Mar-27
Document File: 8 page(s) / 381K

Publishing Venue

IBM

Related People

Abdelnour, GM: AUTHOR [+4]

Abstract

In a bus oriented systems performing management of interconnections from multiple asynchronous attachments with a wide range of speeds, frame sizes, and bandwidth access requirements, a control and access mechanisms is developed to maximize bus throughput, minimize arbitration interference, and insure fairness of packet delivery. Items to be considered are media cycle loss, switching/arbitration overhead, and transfer delay. These factors become even more important when very short packets or frames are presented for transfer and fairness access, delivery, and bounded delays are required without inducing packet loss.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 16% of the total text.

Time Clustered Access for Maximizing Bandwidth Utilization via Widows
of Asynchronous Arrival

      In a bus oriented systems performing management of
interconnections from multiple asynchronous attachments with a wide
range of speeds, frame sizes, and bandwidth access requirements, a
control and access mechanisms is developed to maximize bus
throughput, minimize arbitration interference, and insure fairness of
packet delivery.  Items to be considered are media cycle loss,
switching/arbitration overhead, and transfer delay.  These factors
become even more important when very short packets or frames are
presented for transfer and fairness access, delivery, and bounded
delays are required without inducing packet loss.

      Given a service media (common bus) having multiple attachments
vying for singular access, the time interval between individual
packets including the change time to a new media master must be
considered if the raw capacity of the media is to be maintained.
That is, the changing of media source from one to another must be at
an absolute minimum if the media's capacity is to be realized.  The
notion being to service media access request within a specified time
limit to insure packet access delivery to and through the media with
a minimum loss of bandwidth.  These losses are caused by lost cycles
between masters, lost cycles between packets, and lost cycles
recognizing the access requests.  All of these lost cycles will
result in an increase of the actual media bandwidth required beyond
the aggregate attachment inputs and contribute to a potential input
packet loss.  By bounding the access time to the bus and utilizing
overlapped poll inquiries and bus access grants, such losses due to
overruns at the input attachment queues can be eliminated.

      To achieve this, the input queuing requirements must be
calculated for the worst case input rates to media capacity, and thus
define the input media buffer sizes to insure no input arrival
losses.  Once the capacities are established, a structure must be
defined to insure that the maximum media throughput can be sustained.
The structure herein defines a polling, access, and request
recognition scheme which utilizes time feedback, request look-ahead,
ping-pong table control, and pseudo round-robin access grants.

      This structure combines many aspects of poll recognition and
access regulation to allow service request and packet arrival
overlap, continuous utilization of the media, clustering of short
packets for transmission, elastic service time to accommodate
variable load distribution, and a proportional unload time based on
traffic input rate.  These functions are coordinated within a
structure which includes the attachments, the media, and the arbiter,
Fig. 1 shows function block relationships.  The access fairness is
accomplished by bounding each attachment's access by a time interval,
called "window of access".  This "window of access" interval...