The MPLS WG Archive

[Curtis Villamizar <curtis@workhorse.fictitious.org>]

In message <B99995113B318D44BBE87DC50092EDA90C0D5501@nj7460exch006u.ho.lucent.c
om>, "Busschbach, Peter B (Peter)" writes:
> Curtis,
> 
> Something must have been lost in the translation, because you seem to interpr
> et most of my statements opposite to what I intended. I like ECMP. My email w
> as intended as a request to accept ECMP as a given and accept that it leads t
> o non-deterministic behavior in the network.
> 
> For fear of further slipping away from the compromise that you referred to, I
>  will leave most of your misinterpretations unanswered, but I do have a quest
> ion about one of your statements. Please see in-line.
> 
> Peter


Sorry.  I did very much misread your response.  Hopefully I'll do
better with the clarification below.


> > > I think that ECMP is an issue only for a subset of traffic. 
> > ECMP is not an is
> > > sue for PW-traffic, since VCCV provides a solution. ECMP is 
> > not an issue for 
> > > services with QoS guarantees, where RSVP-TE is used to set 
> > up the connection,
> > >  because an explicit path is nailed down. 
> > 
> > QoS and explicit path are orthogonal.  Some provider may feel that
> > they must combine the two in their service but it is not generally the
> > case.  QoS and ECMP are also orthogonal.  Lets not try to sneak in
> > assertions that may be true in your case but are not generally true.
> 
> I don't understand why RSVP-TE and ECMP are orthogonal. If one uses RSVP-TE t
> o reserve bandwidth between two points, doesn't that nail down an exact end-t
> o-end path? 


First of all:  RSVP/TE != explicit-path

They are related but not equal.  RSVP/TE paths can be determined by
the ingress based on the CSPF computation.  Explicit-path means the
path is predetermined and fixed.  RSVP/TE supports explicit-path as an
option.  When determined by the CSPF (explicit-path is not being
used), the path is in a sense nailed down, at least until the
adaptivity timer fires and a later CSPF decides there is a better path
now.

Also reserving bandwidth != QoS.  In most applications this is an
accounting action only - no queueing is allocated according to the
amount of reservation unless explicitly supported and configured to do
so and most routers don't even offer the ability to set quueuing
according to the bandwidth reservation as an option.  While I
personally think that setting WFQ weights according to the bandwidth
reservation is a good idea (and it appears you do too), ISPs don't
seem very interested in doing this.

There is also no reason why QoS can't exist with LDP and in fact it
does.  The simplest QoS is based on the EXP bits only.  Queueing is
set to favor some EXP values over others.  The amount of traffic with
the preferred EXP values must be limited.  In most cases it is limited
by market acceptance (very few willing to pay extra for higher QoS).

Another issue is whether a MPLS path is a single path.  In LDP,
clearly it need not be - if an ECMP situation exists anywhere along
the path traffic will split if ECMP is enabled.  With RSVP/TE if one
of the hops is a hierarchical hop, then although it is single logical
hop, there may be two or more LSPs configured between the pair of
nodes that are used to split the traffic.  Even if it were an
explicit-path (configured by operator or NMS) if a hop was a
hierarchical hop and ECMP was configured, the traffic would split (if
traffic is such that microflows could be identified).

In some routers a very large logical link between a pair of nodes may
be a concatonated set of lower bandwidth links.  For example, eight
OC48c can form a 20 GB logical link or eight OC192c can form a 80 GB
link (this is not a hypothetical btw).  Since this is a logical link,
it can either be assumed that the pair of nodes can handle monitoring
the link (which has proven safe, but..) or this can be treated as a
form of hierarchy and as an ECMP case to explicitly test all insegment
mappings from a tool such as MPLS Ping.

Without hierarchical LSPs, RSVP/TE does not have multipath unless 1)
more than one LSP is configured to the same destination with equal
cost and MP is enabled, or 2) more than one LSP is on a path to
different egress and the total cost is the same, and this form of MP
is both supported and enabled.  This form of MP is easy to deal with
from an OAM standpoint because the only branch is at ingress.

QoS based on EXP can be enabled and ECMP can also be enabled.  If each
branch of the path honors the EXP bits, QoS still works and exists in
the presense of ECMP, over LDP or RSVP/TE.  There is very clear
existance proof of this.

Curtis