Re: How to enforce BFD to be sent over different paths betweentwosystems?
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: How to enforce BFD to be sent over different paths betweentwosystems?



BFD is suitable for any multipath scenario in which the sessions can be bound to those individual paths by one means or another. This is a somewhat reflexive statement, but it gets at the heart of the problem: if you can't distinguish multiple paths from the vantage point of where BFD is running, you can't have multiple sessions.

This tension is most obvious in the L1/L2 bundling case; if BFD sees only the aggregate interface, there's no use in having multiple sessions (and you can't demultiplex the sessions without some other help, such as out-of-band discriminator exchange.)

One could imagine running BFD on each individual link of a bundle and then having a separate session that runs over the bundle (or not) but it looks like BFD won't get much traction at L2 so this is moot.

In the usual IP ECMP case, BFD is suitable because sessions can be bound to each of the individual L2 paths (by physical or logical interface, for example.)

--Dave

On Sep 23, 2009, at 9:00 PM, Mach Chen wrote:

Correct:-)

"Suitable" should be "not suilable"


--------------------------------------------------
From: "Mach Chen" <mach at huawei.com>
Sent: Thursday, September 24, 2009 10:54 AM
To: "Vishwas Manral" <vishwas.ietf at gmail.com>; "Shahram Davari" <davari at broadcom.com >
Cc: <rtg-bfd at ietf.org>; <dkatz at juniper.net>; <dward at cisco.com>
Subject: Re: How to enforce BFD to be sent over different paths betweentwosystems?

Hi,

For LDP LSP, it also has the same ECMP issue.

Could I say that BFD is suitable for any ECMP scenario?

Best regards,
Mach


--------------------------------------------------
From: "Vishwas Manral" <vishwas.ietf at gmail.com>
Sent: Thursday, September 24, 2009 1:35 AM
To: "Shahram Davari" <davari at broadcom.com>
Cc: <rtg-bfd at ietf.org>; <dkatz at juniper.net>; <dward at cisco.com>
Subject: Re: How to enforce BFD to be sent over different paths between twosystems?

Hi Shahram,

That is correct. For IP there is only one path (shortest path) between two
systems.
Except ofcourse for the ECMP case.

Thanks,
Vishwas

________________________________
From: Linda Dunbar [mailto:ldunbar at huawei.com]
Sent: Wednesday, September 23, 2009 10:10 AM
To: Shahram Davari; rtg-bfd at ietf.org; dkatz at juniper.net; dward at cisco.com
Subject: RE: How to enforce BFD to be sent over different paths between two
systems?

Hi Shahram,



Thank you very much for the answers.

Maybe some wording can be improved. For example, if the text says
“transmitting BFD” over multiple paths, it is better to define what “path” means. To many people, especially people with transport network background, paths mean physical paths. Different LDP or LSP paths between two systems
may not traverse all the paths between two systems.



I understand that the intent of BFD is to let BFD run over various media. But the description of the protocol is under the assumption that BFD is running over a path which source can control, like LSP or LDP. For IP forwarding, the source can’t control which path to traverse from A to B. It
is up to intermediate nodes to choose a path.



Linda Dunbar



________________________________

From: Shahram Davari [mailto:davari at broadcom.com]
Sent: Wednesday, September 23, 2009 11:53 AM
To: Linda Dunbar; rtg-bfd at ietf.org; dkatz at juniper.com; dward at cisco.com
Subject: RE: How to enforce BFD to be sent over different paths between two
systems?



Hi Linda,



I am sure Dave will answer these questions better that I do, but let me give
you my 2c inline.



Regards,

Shahram



________________________________

From: rtg-bfd-bounces at ietf.org [mailto:rtg-bfd-bounces at ietf.org] On Behalf
Of Linda Dunbar
Sent: Wednesday, September 23, 2009 9:35 AM
To: rtg-bfd at ietf.org; dkatz at juniper.com; dward at cisco.com
Subject: How to enforce BFD to be sent over different paths between two
systems?

Dave,



I have some questions on “draft-ietf-bfd-base-09.txt”. Hope you can help.



1. Section 3: 3rd line of the first paragraph states “A pair of systems transmit BFD packets periodically over each path between the two systems”.





Question: when there are multiple paths between two systems, do you mean to have multiple BFD sessions between those two systems, with each session
covering individual path? How to enforce each path being traversed?



SD> For example there can be multiple LSPs between two systems and you need
to run BFD separately on each LSP.



2. The Echo function is pretty much like “Ping”. Each system can
initiate a “Ping” to another system. Is “periodic Ping” an accurate
description of the “Echo function”?



SD> Your understanding is correct. But note that Echo packets are not BFD
packets. BFD just negotiates the Echo interval.





3. Section 4.1 under the “Control Plane Independent” sub- section:

The first paragraph states “if clear, the transmitting system’s BFD
implementation share fate with its control plane”.

Question: When the transmitting system is running multiple routing
protocols, more than one signaling schemes for different services, is it necessary to indicate which routing protocol and which signaling protocol?
  Actually, BFD is to test connectivity which can be up when the
corresponding control plane is done. What is the reason to have BFD share
fate with its transmitting system’s control plane?



SD> Every LSP or PW or tunnel that runs BFD could be setup using an instance of control plane. We don't care about the control plane of the client or server layers. What this bit indicates is the control-plane for the layer
you are running the BFD on.



4. The BFD’s Control Packet Format described in Section 4.1 has a bit field for Demand mode. Why not having a bit field for the other two modes
(Async and Echo)?

SD> If D=0 it means Async mode. For Echo, if a system does not want to receive Echo it can set "Required Min Echo RX Interval " = 0 . And there is
not need to signal that you want to Tx Echo.



5. Is the Discriminator field of the BFD’ Control Packet Format same as unique identifier for particular BFD session from one system? Why not call
it Identifier? Is it negotiated between the two systems?

SD> the Discr is a locally unique number (not globally) very similar to an LSP MPLS label that is distributed from a downstream node. It it not
negotiated.



6.      Section 6.18.17 Concatenated Paths

In transport network, Concatenated paths mean to combine (or bundle) multiple paths to form a bigger path which has higher bandwidth. Therefore,
failure on one of the paths concatenated together will not cause
connectivity problem for the two systems exchanging BFD. This failure will only cause the bandwidth of the concatenated path to be smaller. Do you mean that when one of the paths within a concatenated path fail, the BFD should
indicate this partial failure of the concatenated path?



SD> As far as I know concatenated paths in transport networks mean stitching
two connections such as two LSPs.



7. Editorial: Section 2 Design: 6th line of the first paragraph:
“making it useful in concert with”? Is it a typo?



Thank you very much for helping me.



Best Regards, Linda Dunbar

Advanced Technology Dept, Wireline Networks,

Huawei Technologies, Inc.





Note: Messages sent to this list are the opinions of the senders and do not imply endorsement by the IETF.