ietf
[Top] [All Lists]

Re: [mpls] [Gen-art] review: draft-ietf-mpls-lsp-ping-relay-reply-04

2014-10-22 08:31:36
It would be good to see a revision that clearly spelled out what the
draft was solving, how the initial end-point knew what to create, and
how the responder knew what to use.  It may well be that there is an
effective solution to the problems here.  I look forward to seeing it in
writing.

Yours,
Joel

On 10/22/14, 12:46 AM, Lizhong Jin wrote:
Hi Joel,
The things may not be that bad. You could add a second address (address B in
our example) with K bit set. The address entry with K bit set must be as a
relay node, and could not be skipped.
Section 4.4 should be changed to: Find the first routable address A, and the
first address B with K bit set. If address A is before address B in the
stack, then use address B as the relay address. Otherwise, use address A as
the relay address.
In that case, if A is the private address, the packet will be firstly
relayed to address B. And address A and B belong to one router. Here I
assume one router at least has one routable address for another AS.

Regards
Lizhong

-----Original Message-----
From: Joel M. Halpern [mailto:jmh(_at_)joelhalpern(_dot_)com]
Sent: 2014年10月22日 11:14
To: Lizhong Jin
Cc: gen-art(_at_)ietf(_dot_)org; mpls(_at_)ietf(_dot_)org; 
ietf(_at_)ietf(_dot_)org;
'draft-ietf-mpls-lsp-ping-
relay-reply.all'
Subject: Re: [mpls] [Gen-art] review:
draft-ietf-mpls-lsp-ping-relay-reply-04

ou are saying that this is only for the case where an AS is using public
addresses for its internal numbering, but is not distributing that address
block
externally?

If so, you need to state that very clearly.
I believe a far more common case is one where the numbering is from a
portion of a publicly allocated space, but firewalled.  Which would
produce
the same problem, but would not be amenable to this solution.
And it is well known that many ISPs do internal number assignment from
private blocks.

So what you are now saying is that this draft solves a very small portion
of the
problem?  But it works for that small portion?  If so, at the very least
you
need to be VERY clear about what cases this works for and what cases it
does
not.  And I fear that even if you are clear, it is going to be very
confusing for
folks who are trying to use it.

Yours,
Joel

On 10/21/14, 10:51 PM, Lizhong Jin wrote:
Hi Joel,
I now see your concern. The "private" word in draft is not correct, I
will remove it. The original motivation of "draft-relay-reply" is from
the scenario where IP address distribution is restricted among AS or IGP
area.
And the IP address is not private address. As I know, most deployed
inter-AS or inter-area MPLS LSP is in the network without private IP
address.

Regards
Lizhong


-----Original Message-----
From: Joel M. Halpern [mailto:jmh(_at_)joelhalpern(_dot_)com]
Sent: 2014年10月22日 10:15
To: Lizhong Jin
Cc: gen-art(_at_)ietf(_dot_)org; mpls(_at_)ietf(_dot_)org; 
ietf(_at_)ietf(_dot_)org;
'draft-ietf-mpls-lsp-ping-
relay-reply.all'
Subject: Re: [mpls] [Gen-art] review:
draft-ietf-mpls-lsp-ping-relay-reply-04

The problem is that the original source A, that we are trying to
reach
with a
reply, has an address that appears to the responder X to be routable.
But the destination that is reached by that address is either a black
hole or
some
other entity using the same address.

The reason for the duplication is that, as described in the draft,
the
source
address for A is a private address.  That same address may well be
reachable
according to the routing table at X.  But it won't get to A.

If the problem is something other than private addressing preventing
reachability, it is likely there is still a mistaken routability
problem,
but I can
not illustrate the failure without some other case being described.

Yours,
Joel

On 10/21/14, 10:06 PM, Lizhong Jin wrote:
Inline, thanks.

-----Original Message-----
From: Joel M. Halpern [mailto:jmh(_at_)joelhalpern(_dot_)com]
Sent: 2014年10月22日 0:06
To: lizho(_dot_)jin(_at_)gmail(_dot_)com
Cc: gen-art(_at_)ietf(_dot_)org; mpls(_at_)ietf(_dot_)org; 
ietf(_at_)ietf(_dot_)org;
draft-ietf-mpls-lsp-ping-
relay-reply.all
Subject: Re: [mpls] [Gen-art] review:
draft-ietf-mpls-lsp-ping-relay-reply-04

In line.

On 10/21/14, 10:36 AM, lizho(_dot_)jin(_at_)gmail(_dot_)com wrote:
Hi Joel, see inline below, thanks.

Lizhong


2014.10.21,PM9:30,Joel M. Halpern <jmh(_at_)joelhalpern(_dot_)com>
wrote :

If the process for this draft is to use the top address that can
be reached in the routing table, then there is a significant
probability that the original source address, which is always at
the top of the list, will be used.  As such, the intended problem
will not be solved.
[Lizhong] let me give an example to explain: the source address A
is firstly added to the stack, then a second routable address B
for replying AS is also added. The reply node will not use address
A since it's not routable, then it will use address B. So it will
work and I don't see the problem.

The whole point of this relay mechanism, as I understand it, is to
cope
with
the case when the responder X can not actually reach the source A.
    Now suppose that the packet arrives at X with the Address stack
A, B,
...
X
examines the stack.  The domain of A was numbered using net 10.
The domain of X is numbered using net 10.  A's address is probably
routable
in X's routing table.  The problem is, that routing will not get to
A.  X
examines
the stack, determines that A is "routable", and sends the packet.
This
fails to
meet the goal.
[Lizhong] The source A you are referring is the initiator, right?
The goal of relay mechanism is to reach the initiator. If X is
routable to the initiator (address A), then it is great, other relay
node in the stack will be skipped.
If the source A you are referring is the interface address of one
intermediate node, then I do not understand "routing will not get to
A.  X examines the stack, determines that A is "routable", and sends
the
packet".
Why routing will not get to A, but A is routable?

Regards
Lizhong



Yours,
Joel