From: Andrew Lissitz \(alissitz\) (alissitz@cisco.com)
Date: Sun Nov 06 2005 - 00:31:45 GMT-3
Hello Jongsoo,
I have not labbed this up for you ... so there may be something to
modify or add to your config thoughts, but here goes. Your questions /
my comments:
* Can I use the next hop address instead of interface for "ip route vrf
up 0.0.0.0 <http://0.0.0.0/> 0.0.0.0 <http://0.0.0.0/> fe1" command ?
The answer is yes. Personal preference, I like to use next hop on any
multi-access interface. Once you added the vrf keyword to this command,
you have told the router which routing table to look in. The next hop
address should be found on a connected interface, so this is fine.
* For traffic flow from Hub to Spoke - Once you add a static route on
the PE2 and PE3 routers pointing the CE2 and CE3 networks, the VRF
routing table contains routing information to reach the networks behind
CE2 and CE3. This is only within the PE2 and PE3 VRF route table until
you redist into MPBGP. Within the ipv4 vrf up and vrf down address
families, you will need to redist static. As you mentioned all PEs will
learn this via MPBGP and update their VRF tables as needed.
So after PE1 knows how to reach these networks, it will send the traffic
to the correct PE.
* What are we missing? Well ... in each of these case the CEs only have
one way to go and that is to the PE. For the Hub CE, if it runs static
routing, will it know where to go? A default static should be added
here as well as to all the CEs.
Real world perspective ... (sorry about this) ... The hub router would
likely be running BGP or an IGP with the PE. In this case, the CE would
learn routing information dynamically. In this limited example, the
hub router only has one link to the ISP network... so this is more
simple.
Jongsoo, I did not follow the route-target comments you made. Perhaps
this is from when you labbed this up? No worries ... I am sure in the
lab this up becomes clear.
We are all lacking my man ... especially me!!!!
________________________________
From: Jongsoo [mailto:bstrt2004@gmail.com]
Sent: Friday, November 04, 2005 8:27 PM
To: swm@emanon.com; Andrew Lissitz (alissitz)
Cc: C&S GroupStudy; FORUM
Subject: Re: We've moved to PE-CE routing! (SP-CCIE)
Guys
Thanks for the information. I am obviously lacking some "basics" about
MPLS VPN.
( I realized I have to think multi-dimensionally to figure out routing
behavior among VRFs via MP-BGP and MPLS tags)
Anyway, using the static method explained by Scott, Andrew and Arun, I
came up with more details hoping I get it right.
1) Static routes only
PE1 is RR and all three PEs are in full-mash.
There are two VRFs of up( from spoke to hub) and down( from hub to
spoke) configured in each PE.
CE1-----VRF=down-fe1-PE1-----Pe2-fe2----VRF=Up---CE2 ( 10.2.2.0/24)
|
|
PE3
fe3
|
VRF Up
|
|
CE3( 10.3.3.0/24)
1-1) for a traffic flow from spoke( CE2, CE3) to Hub ( CE1) ( VRF = up)
In PE1 config,
A) put a default static route to CE1 = "ip route vrf up 0.0.0.0 0.0.0.0
fe1" ( fe1 is the out-going interface to CE1)
B) Redistribute the static to bgp, route-export = 100 ( due to
full-mash Mp-ibgp, PE2 and PE3 learn this route),
In PE2 and PE3
A) put the interface connecting CEs to VRF=up
B) Import routes w/ route target =100 from MP-BGP
C) CE2 and CE3 have a default route to PEs
In this way, any traffic from CE2 and CE3 should come into PE1 via MPLS
and go out via the fe1 interface of PE1 to CE1
Can I use the next hop address instead of interface for "ip route vrf up
0.0.0.0 0.0.0.0 fe1" command ?
1-2) for a traffic flow Hub to Spoke ( VRF = down)
In PE2 config,
A) put a static route to CE2 = "ip route vrf down 10.2.2.0 255.255.255.0
fe2" ( fe2 is an out-going interface to CE2)
B) Redistribute the static to bgp, route-export = 200 ( due to
full-mash Mp-ibgp, PE1 and PE3 learn this route but only PE1 imports
it),
In PE3 config,
A) put a static route to CE3 = "ip route vrf down 10.3.3.0 255.255.255.0
fe3" ( fe3 is an out-going interface to CE3)
B) Redistribute the static to bgp, route-export = 200 ( due to
full-mash Mp-ibgp, PE1 and PE2 learn this route only PE1 imports it),
In PE1
A) put the interface connecting CE1 to VRF down
B) Import routes w/ route target =200 from MP-BGP advertised by PE2 and
PE3
C) CE1 have static routes to PE1 for ip prefix destined to CE2 and CE3
In this way, any traffic from CE1 to CE2 or CE2 will be sent to PE1,
which will forward it based on VRF down routing table.
What am I missing?
Jongsoo
On 11/4/05, Scott Morris <swm@emanon.com> wrote:
This is definitely fun to keep the SP list alive, although I'm
surprised we
haven't irritated the R&S guys yet!
Anyway... If I were presented a situation (again, real life
thinking) where
someone said they wanted to essentially set the "hub" site up as
a
route-reflector, my first inclination would be to do static
routes between
the PE and CE and let their CE routers peer with each other and
not with the
PE at all. That would seem to be a simple way of the SP saying
"not my
problem" and save a few headaches.
Like Andrew says, the goal here is to make the CE fairly
ignorant of the
VRF's existance (although with allowas-in, it kinda makes that
moot).
If this were a lab scenario, you'd certainly need to look at
what things you
were or were not allowed to do, but I'd keep it as simple as
possible!
Scott
-----Original Message-----
From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On
Behalf Of
Andrew Lissitz (alissitz)
Sent: Friday, November 04, 2005 12:43 AM
To: Jongsoo
Cc: C&S GroupStudy; FORUM
Subject: RE: "neighbor allowas-in" command ( SP CCIE)
You are right Jongsoo ... I need to keep my head out of reality
for the CCIE
prep ...
As you know, for CCIE lab, so much will depend on wording of
question.
Thanks for the diagram, it helps.
So if CE2 and CE3 must go through CE1, then these two sites must
choose to
route to the hub site first.
Normal routing would work. Forgive me if this is boring anyone
... I am just
going to 'think out loud' in this email.
The CE and the PE share routes. The PE keeps these routes in
the VRF, the
CE is (typically) ignorant of VRF and does not know what goes on
within the
PE. Using iBGP all PEs will share this information so that any
PE with the
same customer VRF will have a converged routing table.
So depending on what you are or are not permitted to do you can
focus on CE
or PE route manipulation. If this is done via the CE, then
manipulate or
filter your IGP so that the next hop is the hub. The PEs simply
share what
they have learned, unless you have configured them otherwise.
If you do this on the PEs, then filter routing information so
the spoke CEs
think that the hub is the next hop for any where. This could be
done either
iBGP, or if the PE and CE run a IGP between them, then filter
the routing
information via the PE IGP. The CE IGP will only learn what is
administratively allowed.
If you can redistribute a static default on the hub PE, cool.
You can allow
only this default to the remotes and allow all | m the spokes to
the hub.
Suggestion to the group... Do you all think we should start a
new email
string to discuss these options? If we are done with this
allowas-in
discussion, then lets create another email string.
Jongsoo, thanks man for keeping these emails going. You going
for your SP
CCIE helps us all!!!
-----Original Message-----
From: Jongsoo [mailto:bstrt2004@gmail.com]
Sent: Thursday, November 03, 2005 8:08 PM
To: Andrew Lissitz (alissitz)
Cc: C&S GroupStudy; FORUM
Subject: Re: "neighbor allowas-in" command ( SP CCIE)
Andrew / All
I totally agree about normal ISP environment. However, in CCIE
lab is far
beyond the normal.
Let me give a scenario
CE1------Pe1-----Pe2--CE2
|
|
Pe3---CE3
CE1 is HQ,
CE2 and CE3 are small office.
For whatever reason( perhaps for virus checking...), CE1 need to
be a hub,
CE2 and CE3 need to be spokes. So that a direct traffic between
CE2 and CE3 is not allowed, it must go through CE1.
I am looking for how many ways exist to design this topology.
Thanks
Jongsoo
On 11/3/05, Andrew Lissitz (alissitz) <alissitz@cisco.com>
wrote:
> Hey Dude,
>
> ISPs run iBGP between PEs (more common to use RRs but same
concept),
> so the idea of a PE seeing it's own ASN in the path is not
likely.
> The PE may see different AS numbers repeated in the path, but
what
> does it care? As long as it does not see its own ASN and
detect a
> loop, all is well.
>
> Some ISPs give the customers one ASN for all the customer
sites. In
> this case, this problem will occur. The CE will see the
routes from
> the remotes sites and see it's ASN in the path. Refer to that
earlier
> diagram.
>
> Other ISPs give the customers different ASN for each site...
but for
> obvious reasons this does not scale.
>
> So, in the case of hub and spoke (is this common, have you all
seen
> this in production?), each PE is running iBGP with each other
or the
> hub is a RR. Each PE will not see it's own ASN. Each PE has
learned
> BGP routes from it's CE and will pass these between peers.
Each peer
> will run the best path process and select the best route.
Only best
> routes are advertised ...
>
> If for some reason, your BGP speakers will see what appears to
be a
> loop, then allowas will work. You can also override the ASN
as well.
>
> Andrew rule of thumb (does not mean much ... will not even get
you
> free
> coffee): allowas-in would be configured on CEs because they
are
> receiving these routes, and as-override would be configured on
routers
> advertising routes ... more than likely the PEs.
>
> Jongsoo, I do not feel as though I have fully answered your
question
...
> Sorry about that.
>
> Andrew
>
>
>
> -----Original Message-----
> From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On
Behalf
> Of Jongsoo
> Sent: Thursday, November 03, 2005 5:06 PM
> To: Andrew Lissitz (alissitz)
> Cc: C&S GroupStudy; FORUM
> Subject: Re: "neighbor allowas-in" command ( SP CCIE)
>
> When I wrote the below email, my stomach was a little empty :)
>
> Regarding my point 2) "advertize iBGP routes to iBGP peers (
normal
> behavior of iBGP is not to advertize iBGP routes to any iBGP
peers)",
>
> First, this thought is from CCO descprition "One Virtual
Private
> Network routing/forwarding instance (VRF) receives prefixes
with ASNs
> from all PE routers and then advertises them to neighboring PE
routers."
>
> I figured from the description that a PE receives a prefix
with its
> own ASN from PE router and advertize
>
> Let's say "neighbor allowas-in 1" in AS 100 in three PE's(
pe1, pe2,
> pe3).
>
> If via MP-BGP, PE1 and PE2 learn from PE3 a VPN route
10.0.0.0/24 with
> as-path = null, I think PE1 will annouce back 10.0.0.0/24 with
> as-path=100 to PE2 and PE3 as well as PE2 will do the same to
PE1 and
> PE3, which is not normal iBGP behavior. But 10.0.0.0/24 with
AS-path
> 100 from PE1 and PE 2 can't beat 10.0.0.0/24 with as-path null
from
PE3.
>
> But the interesting result is if PE2 is not importing
10.0.0.0/24 with
> as-path=null from PE3, then PE2 will only have 10.0.0.0/24
with
> as-path=100 from PE1.
> This will achieve hub-and-spoke VPN topology ( PE1 = hub and
PE2 and
> PE3 = spokes)
>
> Please feel free to correct me if I am wrong.
>
> Jongsoo
>
> On 11/2/05, Andrew Lissitz (alissitz) <alissitz@cisco.com>
wrote:
> > Hey Buddy,
> >
> > Here is a live example, I have not done the hub and spoke
labs like
> > several others on this mail list have:
> >
> > CE ---bgp---PE ---(ISP Cloud)--- PE---bgp---CE
> >
> > Each CE runs AS 65000 and shares routes with the PE. The
PEs share
> > routes via iBGP. The remote PE shares routes with the
remote CE,
> > and the CE sees the routes from AS 65000.
> >
> > What is BGP to do? It sees its own AS number and realizes
there is
> > a problem.
> >
> > Solution: Either PE changes the AS number with as-override
or the CE
> > allows its own AS number to come in via the allowed-as
command. The
> > number @ the end is how many times the CE will allow it's
own AS
> > number to be present in the path string of the incoming
route
> information.
> >
> > Concerning your gut feelings (btw ... hope you are not
writing on
> > empty stomach), number one sounds good, but with number 2,
you are
> > essentially saying that this command will override bgp split
horizon.
>
> > This is not what it will do, if a route is already coming
in, and it
> > contains the BGP's AS number in the path, then let this in.
Not
> > whether or not to advertise to other peers. I have not seen
this
> > command change BGP split horizon behavior ...
> >
> > BGP best path selection still occurs, it is just that the
routes
> > will not be rejected because of loop detection. I have not
seen
> > multiple routes being accepted as best paths... Can multiple
routes
> > exist without the BGP multipath command?
> >
> >
> >
> > -----Original Message-----
> > From: nobody@groupstudy.com [mailto:nobody@groupstudy.com ]
On Behalf
> > Of Jongsoo
> > Sent: Wednesday, November 02, 2005 7:33 PM
> > To: C&S GroupStudy; FORUM
> > Subject: "neighbor allowas-in" command ( SP CCIE)
> >
> > I am trying to understand this command will allow to receive
MP-bgp
> > vpn routes with the same ASN.
> >
> > If I see usage guide in CCO, it says
> >
> > ##################################
> > Usage Guidelines
> > In a hub and spoke configuration, a PE router readvertises
all
> > prefixes containing duplicate autonomous system numbers. Use
the
> > neighbor allowas-in command to configure two VRFs on each PE
router
> > to
>
> > receive and readvertise prefixes are as follows:
> >
> > One Virtual Private Network routing/forwarding instance
(VRF)
> > receives prefixes with ASNs from all PE routers and then
advertises
> > them to neighboring PE routers.
> >
> > The other VRF receives prefixes with ASNs from the customer
edge
> > (CE)
>
> > router and readvertises them to all PE routers in the hub
and spoke
> > configuration.
> >
> > You control the number of times an ASN is advertised by
specifying a
> > number from 1 to 10. "
> > ################################################
> >
> > In my gut feeling, basically, this command seems allow two
things,
> > 1) receive BGP routes with its own ASN from PE or CE, (
normal
> > behavior of BGP blocks BGP route with its own ASN in order
to
> > prevent loop) and
> > 2) advertize iBGP routes to iBGP peers ( normal behavior of
iBGP is
> > not to advertize iBGP routes to any iBGP peers).
> >
> > What seems interesting is this feature will creates a lot of
> > redundant
>
> > routes but the length of AS path will quickly determine the
best
> > routes so that there won't be any loop...
> >
> > This will be a perfect command to make hub and spoke
topology to
> work...
> >
> > The biggest question I have now is " am I right or wrong?".
> > Someone please correct me if I am wrong.
> >
> > Thanks
> >
> >
> > Jongsoo
> >
> >
This archive was generated by hypermail 2.1.4 : Thu Dec 01 2005 - 09:12:05 GMT-3