The MPLS-OPS Archive

[Robert Raszuk <raszuk@cisco.com>]

It not so much matters if your routers are TE capable or not. What
matters is how far are you enxtending your TE tunnels. Some use
core-core some use edge-edge and of course in the latter case FA's seem
not too much applicable.

R.

> "Li, Eve" wrote:
> 
> Hi, all,
> 
> This debate has been interesting. However, can anyone shed a light on the
> practical use of the IGP shortcuts & LSP links feature?
> 
> If a network has all routers that is TE-capable, do we still need this
> feature? It looks like to me the feature is mostly for the staged migration
> from non-TE capable routers to a whole TE aware network.
> 
> -eve
> -----Original Message-----
> From: Robert Raszuk [mailto:raszuk@cisco.com]
> Sent: Wednesday, June 06, 2001 7:02 PM
> To: Alexander Marhold
> Cc: bneal@broadwing.com; mpls-ops@mplsrc.com
> Subject: Re: IGP shortcuts vs LSP Links
> 
> Hi all,
> 
> This thread is getting out of sync a bit. The confusion is caused by the
> fact that one group of people is talking about cisco behaviour and other
> about juniper behaviour.
> 
> What Brad Neal said is quite correct in respect to the juniper
> implementation - what Alex said is mostly correct in respect to cisco
> way :).
> 
> So let me summarize that what juniper calls "IGP shortcut" is the same
> to what cisco calls "autoroute announce". Now there are slight
> differences in the forwarding of traffic into the te tunnels without
> those commands.
> 
> In Juniper all packets with bgp next hop matching the TE tunnel dst will
> be forwarded via the TE (without the IGP shortcuts). In order to also
> forward via the given tunnel packets to dst located behind the tunnel
> tail the "IGP shortcut" needs to be enabled.
> 
> In Cisco by default (without the autoroute announce command under the TE
> tunnel cfg) only what ever static route point to a tunnel will be
> forwarded via such a tunnel. After enabling the "autoroute annouce" CEF
> will be automatically updated with the pointer to tunnel for
> destinations located on the TE tail and behind the tail.
> 
> Now reg the LSP advertisemnt as a link (which is not related feature to
> the above discussion) it is aviable in both cisco & juniper in the same
> way. It is dangerous but under some careful cases useful feature. The
> bi-dir check is done at the SPF. Modifing an IGP cost is quite a
> chalange for dual connected end routers and is not dynamic in case of
> one of the LSP's failure.
> 
> R.
> 
> > Alexander Marhold wrote:
> >
> > I now perfectly understand what you are showing.
> >
> > However my opinion on that is thze following:
> >
> > 1. an LSP tunnel is another labelpath and not a physical real shortcut,
> >  so it uses in the best way the IGP shortest path and in all other ways a
> > worse one (purely from the physical and metric point of view) The reason
> for
> > using another path than the best one is typically based on overloading of
> > certain links and not to find a faster one than the IGP has found ( by the
> > way that is the reason why a tunnel independent of the path per default
> gets
> > the best path metric form tunnel head to tail)
> >
> > 2. so as A does not know about the tunnel, he uses the topology infos and
> > from the "physical" point of view and calculates the best path.
> >
> > So for this example "physically" seen A makes the best possible decision
> >
> > And now lets modify the example a little and change the metric A to C to
> 40
> > in this case A would calculate the best path to be A-B-D-E-F.
> > the tunnel from B to E was set up in order to reduces the traffic on the
> > overutilized link B-D (and for no other reason) as Traffic Engineering
> > should help better utilizing links by intelligently diverting traffic.
> >
> > 1. now A without knowing the tunnel makes the correct decision
> > 2. now B knowing that the tunnel is manually made the shortest path from B
> > to E and to all nodes downstream of E forwards the traffic into the tunnel
> > (in order to prevent overloading of certain links
> >
> > again IMHO the situation of the network is okay and the best decision has
> > been taken.
> >
> > The AUTOROUTE algorithm which is responsible for entering tunnels into the
> > routing table IMHO is quite clever designed, so that the typical GRE and
> > other tunnelproblems of going through the tunnel and then back upstream
> the
> > physical path is typically not the case.
> > It also prevents the wrong assumption of routing protocols that the tunnel
> > is a one-hop and thus automatically preferred, or even prevents the good
> old
> > recursive tunneling problem.
> >
> > Additionally IMHO it would be problematic to announce a unidirectional
> path
> > as a LINK in any LINK_STATE_PROTOCOL, as you would never see a hello from
> > your neighbor received on a unidirectional send-only tunnel and no
> > neighborship means it is not a LINK.
> >
> > So the interesting and not easy to understand issue often is, that A
> without
> > knowing the existence of the tunnel starting at B is using the tunnel if
> his
> > next-hop for a destination is B and if B's next-hop for that destination
> is
> > best reached via the tunnel as the "default" LSP (setup along the IGP best
> > path) is built by independent decisions of each hop along the path.
> >
> > But to be honest, autoroute in all its consequences is not easy to
> > understand, not well documented and it also took me quite some time and
> > effort before I was able to teach that things and answer almost all
> > constructed cases around that issues.
> >
> > with best regards
> >
> > Alexander Marhold
> > Senior Consultant and Trainer
> > http://www.proin.com
> >
> > -----Original Message-----
> > From: bneal@broadwing.com [mailto:bneal@broadwing.com]
> > Sent: Wednesday, June 06, 2001 3:58 PM
> > To: mpls-ops@mplsrc.com
> > Subject: RE: IGP shortcuts vs LSP Links
> >
> > Here's how I understand it:
> >
> >      15---------C------------------15
> >     /          / \                   \
> >    /          /   \                   \
> >   /          10   10                   \
> >  /          /       \                   \
> > A---10----B----10---D----10---E----10----F
> >           |                   |
> >
> >           |                   |
> >
> >           +- - - - LSP (5) - -+
> >
> > (The letters are routers, the numbers are link metrics, and 'LSP' is a
> > TE-tunnel that takes the path B-C-D-E)
> >
> > IGP shortcuts allow B to send IGP traffic destined for F via it's LSP.
> >
> > However, If B does not advertise it's LSP with it's IGP, A will send
> F-bound
> > traffic along the path A-C-F because that path has an end-to-end cost of
> 30.
> > When B injects the LSP into its IGP, A now sees a path A-B-E-F with an
> > end-to-end cost of 25.
> >
> > -----Original Message-----
> > From: Robert Raszuk [mailto:raszuk@cisco.com]
> > Sent: Tuesday, June 05, 2001 5:40 PM
> > To: Ajay Virginkar
> > Cc: mpls-ops@mplsrc.com
> > Subject: Re: IGP shortcuts vs LSP Links
> >
> > Ajay,
> >
> > The way I see them different is that "IGP shortcut" allow the IGP
> > running on the headend of given TE-LSP to use an LSP as the next hop as
> > if it were a sub-interface from the ingress router to the egress router.
> >
> > The concept of advertising LSPs as links on the other hand allow other
> > routers to make a more clufull forwarding decision which may prefer to
> > forward an ipv4 packet to the given LSP's headend.
> >
> > In summary I think those are compelmentary features not exclusive.
> >
> > R.
> >
> > > Ajay Virginkar wrote:
> > >
> > > Let us assume that these are TE-LSPs.
> > > Thanks
> > > Ajay
> > >
> > > ----- Original Message -----
> > > From: "Robert Raszuk" <raszuk@cisco.com>
> > > To: "Ajay Virginkar" <virgink@hotmail.com>
> > > Cc: <mpls-ops@mplsrc.com>
> > > Sent: Tuesday, June 05, 2001 2:33 PM
> > > Subject: Re: IGP shortcuts vs LSP Links
> > >
> > > >
> > > > Before anyone responds pls clarify which MPLS LSPs you have in mind
> > > > TE-LSPs or prefix based LDP build LSPs ?
> > > >
> > > > R.
> > > >
> > > > > Ajay Virginkar wrote:
> > > > >
> > > > > Hi,
> > > > > it seems to me that IGP shortcuts and the concept of using LSPs as
> > > > > links in the IGP serve the same purpose. Both are used for moving
> > > > > traffic destined towards the tail end and downstream routers over
> the
> > > > > tunnel.
> > > > > My question is that are there any advantages / disadvantages of
> using
> > > > > one method over the other.
> > > > > I also assume that these mechanisms can be used only within an Area
> > > > > (OSPF) since we do not have topology information for inter Area
> > > > > destinations.
> > > > > TIA
> > > > > Ajay
> > > > >
> > > > >
> > > > >
> > > >
> >
> > -------
> > 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
> >
> > -------
> > 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

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