************************************************************************

       ATM Forum Document Number: ATM Forum/95-468

       ************************************************************************

       Title:  The Case of Negative ABR Bandwidth:
       Need for a probe interval parameter

       ************************************************************************

       Abstract:
       When all VBR VCs start using their peak cell rate, the total load may
       exceed the link capacity making the capacity available for ABR negative.
       A solution to this problem is described.

       ************************************************************************

       Source:

       Raj Jain, Shiv Kalyanaraman, and Rohit Goyal
       The Ohio State University
       Department of CIS
       Columbus, OH 43210-1277
       Phone: 614-292-3989, Fax: 614-292-2911, Email: Jain@ACM.Org

       The presentation of this contribution at the ATM Forum is sponsored by
       NIST.

       ************************************************************************

       Date: April 9-14, 1995

       ************************************************************************

       Distribution: ATM Forum Technical Working Group Members
       (Traffic Management)

       ************************************************************************

       Notice: This contribution has been prepared to assist the ATM
       Forum. It is offered to the Forum as a basis for discussion and
       is not a binding proposal on the part of any of the contributing
       organizations. The statements are subject to change in form and
       content after further study. Specifically, the contributors
       reserve the right to add to, amend or modify the statements
       contained herein.

       ************************************************************************

       Network links are generally shared by CBR, VBR, and ABR, with CBR
       taking the highest priority, VBR taking the next priority and ABR
       running at the lowest priority. The capacity available to ABR is:

       ABR bandwidth = Link bandwidth - CBR bandwidth - Bandwidth used by VBR

       A similar relationship exists for the allowable input rates:

          Allowable ABR input rate = Link bandwidth - CBR input rate
                               - VBR input rate

       The VBR connections declare an average sustained cell rate (SCR)
        and a peak cell rate (PCR).  The PCR is typically much higher
       than SCR. For example, a VBR connection may declare  SCR and PCR
       of 10 Mbps and 100 Mbps, respectively. The bandwidth used by VBR
       changes continuously resulting in changes to the bandwidth
       available to ABR. The explicit-rate feedback to the sources must
       follow the changes quickly to avoid building up large queues in
       the switches and to avoid cell loss.

       In many situations, the VBR may be overbooked in the sense that
       the sum of PCRs may exceed the link capacity. In such cases, the
       allowable ABR input rate may turn negative. In other words, if
       the switch was allowed to, it would take the buffer space away
       from the ABR and give it to VBR. This is the case of negative (or
       zero) ABR bandwidth.

       When the ABR bandwidth is zero, the ER returned to the sources is
       MCR or zero. If a sources' rate is set to zero, it cannot send
       any further RM cells since they are counted in the sources
       allocated rate.  The next question is how can it start
       transmitting again. We looked at the following two solutions to
       this problem:

       1. Send a "Restart" RM cell from the switch

       2. Sources are allowed to send periodic "probe" RM cells

       After some thought it becomes clear that the first approach will
       result in too much overhead for the switches and in unnecessary
       traffic in the network since all switches may send the restart
       cell to the same source.  We, therefore, prefer the second
       solution which consists of the waiting sources being allowed to
       send RM cells periodically with a CCR field of zero.  Note that a
       waiting source is one which has data to send but has been denied
       transmission due to network overload. It is different from an
       inactive source which have no desire to send any data.


       Current source specs allow inactive sources to send an RM cell if
       they become active after 100ms. Although a waiting source can use
       this mechanism and send an RM cell every 100 ms, the time is
       quite long for a 155 Mbps link and even longer for 622 and higher
       speeds. Our proposal is to allow waiting sources to send the
       probe RM cells every N cell-times, where the cell time is
       computed at PCR of the VC.

       The N should be set small so that the link capacity is used as
       soon as it becomes available. It should be large so that a large
       number of waiting sources can send the probe cells without
       unduely loading the link. A reasonable value is in the range of
       1000 to 3000 cells causing probe cells to be sent every 3ms to
       10ms on 155 Mbps links.

       MOTION:

       Add the following sentence to the source behavior:

       "A source whose ACR has been set to zero due to extreme network
       overload, can send RM cells every PI/PCR seconds, where PI is the
       probe interval parameter specified in number of cells."