The MPLS WG Archive

[suchauhan@hss.hns.com]

David!
     I mostly agree with your points except a few. Please see my 2 pennies
inline ;o).

kind regards,
sumit





"David Waters" <davidwaters@runbox.com> on 09/17/2002 07:08:16 PM

To:   mpls@UU.NET
cc:    (bcc: Sumit Chauhan/HSS)

Subject:  Re: Nodal Fault




Sumit,
IMHO this is not the exact picture. Let me attempt to rephrase it.
The nodal faults and the control channel faults only apply when there are
mechanisms to detect data channel failures independent of the control
channel failures. Because otherwise all our just faults. We in our design
can detect them independently since data faults are recognized by (1) loss
of light or (2) SONET/SDH detects transport layer failure, etc. Nodal
faults then apply when we loose control information but we still maintain
the data plane connections. This can happen in many cases (as mentioned in
the original mail).
#################################>>>>sumit
You are correct in saying that the faults in the data plane and control plane
can be segregated but the point I wanted to make is that the design of the
signalling and the routing protocols (control plane) should be such that they
support it. Considering that the routing or signalling deamon goes down and the
reliable (e.g. TCP session in LDP) is timed out on the other side before the
data plane comes up again, then whether we are retaining the datapath
information or not is immaterial. This is cos' the signalling protocol will
assume that the node has failed and will take corrective action. Do I still miss
something?
##################################>>>>>

What i believe Mareline wants to know is that how other than the examples
which she (i assume!) cited can we have a scenario when we loose the
control plane information but still retain the data plane info.

This is something like the "hitless" concept we have in our routing
protocols wherein we maintain our routing protocol adjacencies if our
daemon goes down till certain time period (assuming that the forwarding
plane is intact!). If the daemon comes up in that time then our neighbours
never come to know of that flap and if it doesnt then we inform everybody
about it!
#################################>>>>sumit
What is critical in the hitless concept is that the timeout of the session
between control plane peers is longer than the time the deamon takes to come up
again after failure. If not then essentially the control plane failure will be
treated (by the peer deamons) as a node failure and
corrective/restoration/backup action will be taken.
##################################>>>>>
The chances of this happening are fairly high in distributed architectures
where the control process is running on a separate processor and the
forwarding on the other (the NPs!)

just my 2 cents.

----- Original Message -----
From: <suchauhan@hss.hns.com>
To: "Mareline Sheldon" <marelines@yahoo.com>
Cc: <mpls@UU.NET>
Sent: Tuesday, September 17, 2002 4:44 PM
Subject: Re: Nodal Fault


|
|
| hi Mareline!
|      IMHO Nodal faults is a genre of faults that include both the data
and
| control path failure. The case of just the control path failing shall be
a
| special case of a Nodal fault.
|
|      Generally the signalling protocols are so designed that they should
come to
| know immediately of the node (control plane) going down and take
corrective
| action. Whether the data plane is up or not is I guess irrelevant for the
| corrective action as     there is no reliable means to determine this (as
the
| control path is down).
|
| best regards,
| sumit
|
|
|
|
| Mareline Sheldon <marelines@yahoo.com> on 09/17/2002 02:49:46 PM
|
| To:   mpls@UU.NET
| cc:    (bcc: Sumit Chauhan/HSS)
|
| Subject:  Nodal Fault
|
|
|
|
| Hi,
| As i understand, the Nodal faults belong to the genre of faults wherein
the
| control plane at
| the node fails while all the data plane connections are maintained. This
could
| be if the
| reachability to the neighbouring nodes is lost, or the control processor
fails
| or after a
| restart or RSVP daemon crashes.
|
| Can there be any other scenario when this may happen i.e control plane
| information is lost but
| data plane is still preserved.
|
| regards,
| mareline s.
|
|
|
| __________________________________________________
| Do you Yahoo!?
| Yahoo! News - Today's headlines
| http://news.yahoo.com
|
|
|
|