The MPLS-OPS Archive

Cell Relay Retreat>MPLS-OPS Archive>month:2003-Feb> msg00139



[Date Prev][Date Next][Thread Prev][Thread Next]  
  [Date Index][Thread Index][Author Index][Subject Index]

Re: Congested TCP/LDP - How to confirm that the label actually made it to the (targeted) peer!

  • From: "wanglei" <wanglei@harbournetworks.com>
  • Date: Thu, 20 Feb 2003 08:48:26 +0800
  • Resent-Date: Wed, 19 Feb 2003 21:03:32 -0500
  • To: "Manish Gupta" <manish@xebeo.com>, <mpls-ops@mplsrc.com>
  • X-MIME-Autoconverted: from base64 to 8bit by host.secure4-hosting.net id h1K0nX505610


in LDP specs, there is no confirm when LSR-B receives label mapping from LSR-A.
in fact ,if the A hasn't receive the label mapping from B, the packet  from A won't
be sent to B with label, it is just sent out as ip packet. 
so ,when A has sent the label mapping , it can't indicate that the LSP is established.
you can see one of the L2VPN protocols (martini), it explains the problem more clearly.


----- Original Message ----- 
From: "Manish Gupta" <manish@xebeo.com>
To: <mpls-ops@mplsrc.com>
Sent: Thursday, February 20, 2003 4:46 AM
Subject: [MPLS-OPS]: Congested TCP/LDP - How to confirm that the label actually made it to the (targeted) peer!


> Hello
> 
>     I have one doubt which I am not being able to resolve by reading
>     LDP specs. Hoping that somebody on the list will be able to help
>     me out.
> 
>     Here is the scenario:
> 
>    [LSR-A] ----- <targeted ldp session> ----- [LSR-B]
> 
>    note.1: LSR-A and LSR-B are *NOT* connected directly.
> 
>    Time-1: <A> and <B> establish a targeted ldp session.
> 
>    Time-2: <A> sends a label map to <B>.
> 
>    Time-3: <B> receives label map from <A>
> 
>    In the above scenario, <A> has no means to detect when did
>    <B> actually received its label. Once <A> puts the packet on
>    the wire, it assumes (thanks to reliable TCP) that label will
>    reach <B>. The difference between Time-3 and Time-2
>    could be significant if network connecting <A> and <B> is
>    congested. The packet may end up sitting in TCP queues
>    for sometime.
> 
>    If I am not mistaken, <A> has no means to detect this time
>    difference and may (incorrectly) start sending out labeled
>    packets to <B> and if data packets reach <B> before label
>    could be installed, they will get dropped (data packet
>    does not have to take the same path as the label pdu).
> 
>    Can anybody explain how one can handle this? Build some
>    sort of end-to-end LDP level reliability.
> 
>    thanks
>    manish
> 
> -------
> The MPLS-OPS Mailing List
> Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
> Archive: http://www.mplsrc.com/mpls-ops_archive.shtml
> 

-------
The MPLS-OPS Mailing List
Subscribe/Unsubscribe:  http://www.mplsrc.com/mplsops.shtml
Archive: http://www.mplsrc.com/mpls-ops_archive.shtml