Browse Prior Art Database

On the Fly Bandwidth Reservation Control

IP.com Disclosure Number: IPCOM000119054D
Original Publication Date: 1997-Nov-01
Included in the Prior Art Database: 2005-Apr-01
Document File: 4 page(s) / 101K

Publishing Venue

IBM

Related People

Brassier, R: AUTHOR [+4]

Abstract

Multiprotocol networks with statistical multiplexing now create the need for a facility to monitor and control the bandwidth occupation. Those networks happen to be high speed networks like Frame Relay, for which both the node processor utilization and latency are critical; a bandwidth control system must, thus, be very efficient instruction-wise and operate with minimal queueing to limit the delay introduced in the data propagation.

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

On the Fly Bandwidth Reservation Control

      Multiprotocol networks with statistical multiplexing now
create the need for a facility to monitor and control the bandwidth
occupation.  Those networks happen to be high speed networks like
Frame Relay, for which both the node processor utilization and
latency are critical; a bandwidth control system must, thus, be very
efficient instruction-wise and operate with minimal queueing to limit
the delay introduced in the data propagation.

      There  are two aspects to be controlled, the throughput and its
distribution between the users (protocols, channels).  An analogy
with light would be luminosity and color.  This disclosure addresses
the latter only.  Consequently, the question is how to enforce a
distribution between protocols competing for access to a shared media
with a statistical input.

      Basically,  the way of controlling bandwidth distribution is to
split the traffic into several groups or classes.  To each class is
associated a ratio versus the other groups.  When, and only when,
there is a congestion of the media (i.e., the available bandwidth is
totally used) does the control take place in order to pace down those
who exceed their ratio and eventually every class, so the ratio now
fits into the available bandwidth that is controlled by some other
mechanism.  A sample is defined so that the ratio applies to a given
amount of traffic.  There are two classical solutions for this:
  1.  Round robin
      In this solution, a separate queue is associated to each
       class.  A round robin selector scans the different queues
       and transmits up to the class share from each queue.  The
       more data in each queue, the better the output distribution
       because the full share of a class as a chance to be sitting
       in its queue; but also, the worst the delay introduced by
       the system and the memory used for it.  Another problem in
       this implementation is the processor instructions wasted in
       polling queues that have no traffic pending.
  2.  Decantation
      This improved solution builds a chain of the class queues
       that are used at a given time.  This saves the polling time
       of unused queues.  A class queue is inserted in the chain
       of queues according to its ratio so the output fits
       expectations.  This insertion mechanism is time consuming,
       making this algorithm inefficient CPU wise.  Worst, the
       main problem of the previous implementation is not fixed.
       As the working sample against which...