Cell Relay Archive

[Paul Koning <pkoning@xedia.com>]

Richard S. Tsumaki wrote:
> 
> Various sources seem to be giving me seemingly conflicting information
> about how UBR is implemented.  Can someone straighten me out by telling
> which of the following statements are false (or true)?  (My primary
> concern is IP traffic implemented across different ATM interface rates -
> how does the UNI/NNI negotiate the rate if at all?):
> · PC based LANE or Classical IP implementations today use UBR (or
> proprietary ABR).

True for LANE.  I think IPOA is less picky.

> · In most implementations there is no CAC for UBR traffic.

Don't know about that.  I suspect a more accurate statement is
that in most implementations CAC has little to do for UBR (since
there are few reasons to reject a call).

> · If there is a CAC for UBR, the CAC function knows the entire network
> topology (how did it get to know the topology?).

You need to distinguish two things that sound very similar but are
very different:

CAC -- a local function (not standardized) where a switch decides
whether IT has the resources needed to support a new call with the
specified parameters.  If it does, it accepts the call and propagates
the setup message.  If not, it rejects it or (if allowed by the setup
message) adjusts some of the parameters so it can support that call.

GCAC ("Generic CAC") -- a function performed at a "DTL Originator"
in PNNI.  That's primarily the first switch in the path (the one that
computes the route), and in addition any entry border node (in the
hierarchical case).  The simplest way to think about GCAC is as
an attempt in the originating switch to predict what the outcome of
CAC will be at each subsequent switch in the route it is considering.
To do so, it uses the advertised node and link parameters from the 
PNNI routing database.  The intent is that the values advertised
should be such that GCAC indeed is a decent predictor of what your
CAC will do.

GCAC of course cannot be fully accurate, for several reasons: (1) it
is a generic approximation of an implementation-specific algorithm,
(2) it can only take into account parameters that PNNI can advertise,
(3) it is based on currently know (i.e., recently advertised) values,
not the state of the other node when the setup will reach it.

To handle the fact that GCAC will sometimes declare a route to be
fine when in fact it isn't, PNNI has the "crankback" mechanism -- if
the CAC in a switch doesn't like the call, it rejects it with a 
message that means "I don't like this but someone else might, so try
another route if you wish".

> · If the CAC function doesn't know the entire network topology, it asks
> at connection setup (what signaling message is that?).

That doesn't compute.  GCAC always uses what's in the topology database.
(That can be abstracted due to hierarchy, of course.)  Perhaps someone
was thinking about the ability to negotiate parameters (with the
Alternative and the Minimum traffic parameters information elements
in the setup message).

> · UBR specifies a PCR and CDVT, but in almost all implementations that's
> just set to the constraints of the physical interface for any VCC.

I'd expect PCR generally to be set to the smallest link rate of any
of the links in the path.  That's of course up to the switches, but
that would work best.

> · Barring any large buffers or packet discard schemes, UBR traffic from
> a high bit-rate source will suffer a high cell loss, if sent to a low
> bit-rate destination.  Any controls on the bit-rate are from the upper
> layer protocols/application at the ends of the connection.

Correct.  That's a good reason to do packet based discard, of course,
and
most switches support it.

	paul

-- 
!-----------------------------------------------------------------------
! Paul Koning, NI1D, C-24183
! Xedia Corporation, 119 Russell Street, Littleton, MA 01460, USA
! phone: +1 978 952 6000 ext 115, fax: +1 978 952 6090
! email: pkoning@xedia.com
! Pgp:   27 81 A9 73 A6 0B B3 BE 18 A3 BF DD 1A 59 51 75
!-----------------------------------------------------------------------
! "The only purpose for which power can be rightfully exercised over 
!  any member of a civilized community, against his will, is to prevent
!  harm to others.  His own good, either physical or moral, is not
!  a sufficient warrant."    -- John Stuart Mill, "On Liberty" 1859