[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: (Correct ver.) Avoiding misconnections in restoration signaling [Was: Protection obj. in restoration signaling]
vishal, see in-line:
In your exchange last month, the following text was proposed as
a way of activating the backup path in the 1:1/Shared Mesh case
to avoid misconnections:
< 02/04/04 Dimitri Papadimitriou wrote:>
by re-reading it was also my understanding when performing the xc
on the resv we got here a transient issue that may appear to be
problematic as the length of the path in terms of number of hops
increases (since the latency increases, the probability to be
impacted by this also increases), so would the following text
address your concern:
"In certain circumstances, it MAY be desirable to perform the
activation of the secondary LSP in the upstream direction (Resv
trigger message) instead of using the default downstream method.
In this case, and in order to avoid any mis-ordering and any mis-
interpretation between a refresh Resv and a trigger Resv message
at intermediate nodes along the secondary LSP, upon reception of
the Path message, the egress node MAY include the PROTECTION
object in the Resv message. The latter is then processed on a hop
by hop basis to activate the secondary LSP until reaching the
ingress node. The PROTECTION object included in the Path message
MUST be set as specified in Section 8.3 and Section 9.3. The
upstream activation behavior SHOULD be configurable on a local
basis."
However, after looking at the exchange in detail (and keeping
in mind the hard/soft preemption issue), I believe the
text needs to be somewhat more precise.
The way it is worded at present (even though it satisfied Yoshihiko),
would appear to allow for the possibility of misconnections, just in
reverse,
as I explain below.
it was assumed as inferred here below (from the
discussion it was clear, otherwise there would have
been no reason to do this)
note as this was part of the discussion we had w/
kohei concerning the detailed procedure - but since
it seems that additional details are in expected in
this document adding a paragraph along these lines
will be done
The text, as worded, only provides for the activation of the secondary
LSP upon receipt of a Resv message for the secondary LSP (which would
be identified by the inclusion of the PROTECTION obj). However, it
makes an implicit assumption that the state for the extra-LSP, in both
the _control and the data_ plane, is brought down during the transmission
of the Path message for the secondary LSP.
Given the debates on the hard/soft-preemption issue, it is not clear
that the state of the extra-LSP will be brought down upon the receipt
of the Path message for the secondary LSP. If that is not
the case, then a misconnection happens as follows.
A-----------B
\ /
C-------D
/ \
E F
A-B: Primary LSP
A-C-D-B: Secondary LSP
E-C-D-F: Extra (preemptable) LSP
Node D, upon receipt of the Resv for the secondary LSP, brings down state
for the extra-LSP, and activates its cross-connect to switch from C-D ->
D-F to C-D -> D-B. However, since node C may not have done so, one again
has a misconnection with traffic flowing from E to B, instead of A to B.
So, it would be useful to have some text that makes explicit that
an intermediate node along the route of a secondary LSP must remove
the state, in the control and data planes, associated with the extra-LSP
(that uses resources required by the secondary LSP) before forwarding
the Path message for the secondary LSP.
a paragraph along the above lines will be added
A few more editorial comments:
i) If the text is under 1:1/Shared Mesh, why "under certain circumstances"
Haven't we established that the behavior suggested by Yoshihiko is needed
under _all_ circumstances.
because it is a trade-off between recovery speed
(also, nothing says that an operator must use the
spare capacity) and resource efficiency usage -
and there no reason to mandate one behaviour or
another - cf. our previous discussion where you
are doing the policing for the operator instead
of defining the set of tools it can use
now of example of circumstances can be further
detailed as long as they are not constraining
ii) The phrase
"and in order to avoid any mis-ordering and any mis-
interpretation between a refresh Resv and a trigger Resv message
at intermediate nodes along the secondary LSP, upon reception of
the Path message, the egress node MAY include the PROTECTION
object in the Resv message.
is unduly long and complicated. It should be broken up into smaller
sentences. Moreover, it is not always clear what
"refresh Resv" and "trigger Resv" refer to. Isn't a "refresh Resv"
a Resv corresponding to the Extra LSP and a "trigger Resv" a
Resv corresponding to the Secondary LSP?
will clarify,
thanks,
- dimitri.
Thanks,
-Vishal
-----Original Message-----
From: owner-ccamp@ops.ietf.org [mailto:owner-ccamp@ops.ietf.org]On
Behalf Of Dimitri.Papadimitriou@alcatel.be
Sent: Wednesday, February 04, 2004 5:03 PM
To: Yoshihiko SUEMURA; ccamp@ops.ietf.org
Subject: Re: Protection object in resv message
hi, see in-line
Yoshihiko SUEMURA wrote:
Hi Dimitri,
Please see inline.
On Sun, 01 Feb 2004 15:39:33 +0100,
Dimitri.Papadimitriou@alcatel.be wrote:
hi, see in-line
Yoshihiko SUEMURA wrote:
P&R Design Team,
In the 1:1/Shared Mesh Restoration described in
draft-lang-ccamp-gmpls-recovery-e2e-signaling-02, activation of a
secondary LSP is done only by a Path message. The Protection object is
carried only in a Path message.
However, I think a Resv message should also carry the
Protection object.
Consider the following case.
A-----------B
\ /
C-------D
/ \
E F
A-B: Primary LSP
A-C-D-B: Secondary LSP
E-C-D-F: Extra (preemptable) LSP
Activating the Secondary LSP using only a Path message may cause
unintended connection (A-C-D-F) between the Secondary LSP and the Extra
LSP.
here i would agree that there is a condition on the next_hop
to delete the extra_lsp state before activating the xc for
the secondary lsp and in order to guarantee this commit of
the resources activation may be done upon resv reception
also the use of hard preemption before committing this
operation decreases (if not completely elevates if used
to commit action when received from D in this example)
the time occurrence of this transient state:
- PathErr with PAth_State_Removed flag towards E and a PathTear
to the destination F
- or a PathErr with Path_State_Removed flag towards F and a
PathTear towards E
therefore there are other faster triggers for this purpose
the issue being at the end to either perform this operation
as fast as possible when reaching the last common node,
or simply delete in downstream direction and commit along
the upstream direction as said above (there are more complex
cases where this might be at the end more easy to process)
This can be prevented by applying a two-way activation scheme using
both Path and Resv messages.
nothing prevent this from the above (the paragraph that
describes this doesn't say commit at the data plane this
is left out to the implementation) some clarification in
the document are certainly needed here
You can delete the Extra LSP by the Path
message, and activate the Secondary LSP by the Resv message.
you may want to apply this activation scheme, in such a case
all the nodes would have their extra-traffic lsp deleted
through the downstream path message
Yes. This is what I want to apply. I want to delete all the
extra-traffic LSPs through the downstream Path message, and then,
activate the secondary LSP through the upstream Resv message.
However, if the Resv message for activation does not carry the
Protection object, it cannot be distinguished from a refresh Resv
message. This still causes unintended connection in the following case.
(1) At node C, a crossconnect for the Extra LSP is deleted when
receiving a Path message.
(2) Then, if node C receives a refresh Resv message from D, it
sets up a
crossconnect for the Secondary LSP because it cannot distinguish the
refresh Resv message from a Resv message for activation.
referring to 2961 p12/13 don't see how see this could happen,
would you clarify, in order to address this point in case, also
the resv is considered implicitly here as trigger message
After (1), node C waits for the upstream Resv message for activating the
secondary LSP. However, it may receive a refresh Resv message (refresh
for a Resv message for PROVISIONING the secondary LSP) from D before
receiving the Resv for activation. Currently, C cannot distinguish it
from the Resv for activation because there is no difference between
their formats. This may trigger C to activate the secondary LSP
unintentionally before the downstream nodes delete their extra-traffic
LSPs.
by re-reading it was also my understanding when performing the xc
on the resv we got here a transient issue that may appear to be
problematic as the length of the path in terms of number of hops
increases (since the latency increases, the probability to be
impacted by this also increases), so would the following text
address your concern:
"In certain circumstances, it MAY be desirable to perform the
activation of the secondary LSP in the upstream direction (Resv
trigger message) instead of using the default downstream method.
In this case, and in order to avoid any mis-ordering and any mis-
interpretation between a refresh Resv and a trigger Resv message
at intermediate nodes along the secondary LSP, upon reception of
the Path message, the egress node MAY include the PROTECTION
object in the Resv message. The latter is then processed on a hop
by hop basis to activate the secondary LSP until reaching the
ingress node. The PROTECTION object included in the Path message
MUST be set as specified in Section 8.3 and Section 9.3. The
upstream activation behavior SHOULD be configurable on a local
basis."
hope this addresses the concern,
thanks,
- dimitri.
Thank you,
Yoshihiko
If (2) occurs before D deletes a crossconnect for the Extra LSP, it
causes unintended connection between the Secondary LSP and the
Extra LSP.
As a solution for the above problem, I propose to add the Protection
object to a Resv message. The unintended connection can be prevented
because you can distinguish a Resv message for activation from
a refresh
Resv message by watching the S bit.
suggest to first clarify the previous point,
thanks,
- dimitri.
Best regards,
Yoshihiko
-----------------------------------------------------------------
Yoshihiko SUEMURA
NEC Corporation
E-mail: y-suemura@bp.jp.nec.com
--
Papadimitriou Dimitri
E-mail : dimitri.papadimitriou@alcatel.be
E-mail : dpapadimitriou@psg.com
Webpage: http://psg.com/~dpapadimitriou/
Address: Fr. Wellesplein 1, B-2018 Antwerpen, Belgium
Phone : +32 3 240-8491
-----------------------------------------------------------------
Yoshihiko SUEMURA
NEC Corporation
E-mail: y-suemura@bp.jp.nec.com
--
Papadimitriou Dimitri
E-mail : dimitri.papadimitriou@alcatel.be
E-mail : dpapadimitriou@psg.com
Webpage: http://psg.com/~dpapadimitriou/
Address: Fr. Wellesplein 1, B-2018 Antwerpen, Belgium
Phone : +32 3 240-8491
--
Papadimitriou Dimitri
E-mail : dimitri.papadimitriou@alcatel.be
E-mail : dpapadimitriou@psg.com
Webpage: http://psg.com/~dpapadimitriou/
Address: Fr. Wellesplein 1, B-2018 Antwerpen, Belgium
Phone : +32 3 240-8491