ATM Switch Interoperability

Steve Murray steve.murray at
Tue Jan 19 17:50:22 UTC 1999

Here is the function for the relationship of Burst Tolerance BT (time) and Maximum Burst Size MBS 
(a number of cells).  Most switches allow you to set the MBS and then they calculate the BT based
on this function.  Definitions: Ts =  1/SCR, Tr = 1/PCR, ts = BT

	MBS = | 1 + ((CDVT + ts) / (Ts - Tr)) |   where  |x| stands for the largest integer less than x.
	for Tr > 1 and CDVT > Tr -1.

This function is based on using a unit of micro-seconds for each variable.  The corresponding function is used
to calculate the maximum burst size B (a number of cells) on the PCR policing:

	B = | 1 + ((CDVT) / (Tr - 1)) |       For Tr > 1  where  |x| stands for the largest integer less         
	than x. CDVT must be greater than Tr - 1.

This function is also based on units of micro-seconds.  It breaks down if Tr < 1 microsecond.  At that point a smaller
unit of time would need to be subtracted from Tr instead of 1 microsecond.  This only matters with very large PVCs
with a PCR of over 1000000 cells per second.

These values for MBS and CDVT are usually defaulted for the user and the same default applies no matter what
the SCR and PCR values are.  Some ATM providers set the CDVT and MBS very low and adjust up if problems like
you are observing appear.

If the lightstream does traffic shape properly this should not be an issue.  If you do have traffic shaping on the lighstream
you could set the PCR=SCR.  You will lose a small amount of burst but will be able to sustain at SCR.  The bursts do not
really buy you a large amount of extra traffic anyway (unless CDVT and MBS are set high) they are really just there to
allow for variations in accuracy of policing between devices.

Steve Murray
MCI Worldcom
steve.murray at

>>> "Martin, Christian" <CMartin at> 01/15 9:19 PM >>>
>We have it matched to the policing policy of the Network Provider.  By
>increasing the CDVT, we gain some bucket depth, but we are only MUXing 5
>VCCs over one VP.  There shouldn't be too much clumping, espescially
>considering the load on the switches(low).  One thing is interesting,
>though:  The far end (customers) are all running over DS3 using HEC
>A curiosity:  Given the parameters below for the contract, particularly
>the MBS and CDVT, does it appear that we are being cheated?  From my
>understanding of the UNI 3.1 spec, the MBS should be adjustable
>according to SCR/PCR, given the fact that CDVT is defined as the same
>for both SCR and PCR buckets.  
>One obstacle that myself and my colleagues have not been able to
>overcome is the use (aside from bucket calculation) of the Burst
>Tolerance.  It is obviously proportional to the MBS, but again, it
>appears as if a contract specifying a PCR/SCR of 100000/50000 cells per
>sec, should have a higher MBS than a 10000/5000 cell/sec contract.
>Any insights?


More information about the NANOG mailing list