[c-nsp] (no subject)
gladston at br.ibm.com
gladston at br.ibm.com
Sun Feb 5 20:21:13 EST 2006
Thanks for the answers.
The value 1514 bytes is: 1472 (data) + 8 (ICMP header) + 20 (IP header) +
14 (ethernet header).
There is 802.1q between two switches, but I am considering that the
switches hands correctly 1518 bytes by default when configured for trunk.
Any different opinion?
This is the net:
Telecom-device-------sw1---------7609(a)---(pos)-----7609(b)--------sw2--------Telecom-device
| |
|____(giga)____|
Telecom-device generates frames with 1514 bytes and DF set on a specific
situation.
-sw1 adds 4 bytes vlan tag.
-7609 removes 4 bytes vlan tag and adds MPLS tag. FRR is also configured,
so when the main connection fails it is added FRR tag. On this case
7609(b) giga interface needs to accept 1514 + 1 MPLS tag + 1 FRR tag
There is a sligth variation of this net, where VPN MPLS is configured, so
7609(b) needs to accept 1514 + 1 MPLS tag + 1 FRR tag + 1 VPN MPLS tag.
What I am missing is the concept of Mark book: "RSVP signaled TE label +
TDP/LDP signaled IGP label).
If I undertood correctly the answers, on PEs there will exit 1 MPLS tag +
1 FRR tag + 1 VPN MPLS tag. On P there will be IGP label. as well. Is that
right?
And about MTU, it is necessary add support for jumbo frame on 7609(b) and
also add the command "mpls mtu" on 7609(a) and 7609(b). Is that correctly?
Cordially,
------------------------------------------------------------------
Alaerte Gladston Vidali
IBM Global Services - SO
Tel.55+11+2121-2879 Fax:55+11+2121-2449
----- Forwarded by Alaerte Gladston Vidali/Brazil/IBM on 05-02-2006 23:04
-----
"Oliver Boehmer \(oboehmer\)" <oboehmer at cisco.com>
05-02-2006 06:41
To
Alaerte Gladston Vidali/Brazil/IBM at IBMBR, <cisco-nsp at puck.nether.net>
cc
Subject
RE: [c-nsp] Number of MPLS labels and MTU
gladston at br.ibm.com <> wrote on Sunday, February 05, 2006 4:17 AM:
> Hi,
>
> Reading Troubleshooting Virtual Private Networks, by Mark Lewis, he
> states that if using VPN MPLS + TE, a label stack depth of three is
> used (RSVP signaled TE label + TDP/LDP signaled IGP label + VPN
> label).
Right, but this also depends on where the tunnel is (for a PE-PE tunnel,
you might not have an IGP label), and if you are using TE Fast-Reroute,
you could have an addtl. label during the backup case.
> Considering an application generates a packets with higher number of
> bytes (1472) and DF bit set, the packet will have 1514 bytes without
> label and 1514 + 4 + 4 +4 bytes travelling through the network. Do
> you agree?
How do you arrive at 1514?
> Would it be enough to configure jumbo frame support and mpls mtu
> under the physical interface or it would be necessary any command
> under the TE tunnel interface?
You need to make sure your physical interfaces will be able to transport
the maximum size (i.e. Payload + 12 (or 16) bytes MPLS shim header).
Setting the MTU on the TE Tunnel interface is not an option (I think)
since you might also sending tagged packets over the tunnel which can't
be fragmented.
oli
=========================================================================
----- Forwarded by Alaerte Gladston Vidali/Brazil/IBM on 05-02-2006 23:04
-----
"Piotr Marecki" <peter at mareccy.org>
05-02-2006 08:29
To
<cisco-nsp at puck.nether.net>, Alaerte Gladston Vidali/Brazil/IBM at IBMBR
cc
Subject
Re: [c-nsp] Number of MPLS labels and MTU
> Hi,
>
> Reading Troubleshooting Virtual Private Networks, by Mark Lewis, he
states
> that if using VPN MPLS + TE, a label stack depth of three is used (RSVP
> signaled TE label + TDP/LDP signaled IGP label + VPN label).
>
> Considering an application generates a packets with higher number of
bytes
> (1472) and DF bit set, the packet will have 1514 bytes without label and
> 1514 + 4 + 4 +4 bytes travelling through the network. Do you agree?
I quess you are talking here about entire L2 PDU size on core ethernet (
without 802.1q ) .
If so , 1526 is really minimum that would not be enough if you enable
other
features
, like 802.1q subinterfaces , FRR , not to mention L2 PWE transport.
>
> Would it be enough to configure jumbo frame support and mpls mtu under
the
> physical interface or it would be necessary any command under the TE
> tunnel interface?
>
You need only mpls mtu command on your core interfaces ( in this example
1512 ) , additional
mtu size increase on swicthes ( if any ) . There is no need to increase
tunnel mtu .
regards
Piotr Marecki
======================================================================
----- Forwarded by Alaerte Gladston Vidali/Brazil/IBM on 05-02-2006 23:05
-----
"Mark Lewis" <mark at mjlnet.com>
05-02-2006 21:21
Please respond to
mark at mjlnet.com
To
cisco-nsp at puck.nether.net
cc
Alaerte Gladston Vidali/Brazil/IBM at IBMBR
Subject
Fwd: RE: [c-nsp] Number of MPLS labels and MTU
Hi,
>
>gladston at br.ibm.com <> wrote on Sunday, February 05, 2006 4:17 AM:
>
> > Hi,
> >
> > Reading Troubleshooting Virtual Private Networks, by Mark Lewis, he
> > states that if using VPN MPLS + TE, a label stack depth of three is
> > used (RSVP signaled TE label + TDP/LDP signaled IGP label + VPN
> > label).
>
>Right, but this also depends on where the tunnel is (for a PE-PE tunnel,
>you might not have an IGP label), and if you are using TE Fast-Reroute,
>you could have an addtl. label during the backup case.
Yep- just as you says, it depends on the location of the tunnel endpoints,
whether FRR is configured, etc- and this is stated and a number of
examples given in the book. And there's also a case study that describes
the label stack of L3VPN over TE tunnels.
>
> > Considering an application generates a packets with higher number of
> > bytes (1472) and DF bit set, the packet will have 1514 bytes without
> > label and 1514 + 4 + 4 +4 bytes travelling through the network. Do
> > you agree?
>
>How do you arrive at 1514?
>
> > Would it be enough to configure jumbo frame support and mpls mtu
> > under the physical interface or it would be necessary any command
> > under the TE tunnel interface?
>
>You need to make sure your physical interfaces will be able to transport
>the maximum size (i.e. Payload + 12 (or 16) bytes MPLS shim header).
>Setting the MTU on the TE Tunnel interface is not an option (I think)
>since you might also sending tagged packets over the tunnel which can't
>be fragmented.
Agreed- you just need to work out the maximum number of labels in the
stack, add this to the max IP packet size that you receive from customers
(assuming L3VPN), and then set this on LSR i/fs using 'mpls mtu' (and
config jumbo frame support on switches). If you're not sure of the label
stack size for your particular config then it may be a good idea to err on
the safe side and add enough 'headroom' for one or two 'extra' labels, if
possible.
Also, note the relatively recent very slight modification to the use of
the 'mpls mtu' command:
http://www.cisco.com/en/US/products/ps6566/products_feature_guide09186a00805b5519.html
Finally, if you are transporting L2 frames over your MPLS backbone then
make sure you take into account the MPLS MTU required for these typically
larger labelled packet sizes (L2 frame + control word + label stack). But
that's covered in detail in chapter 7 :)
Cheers,
Mark
More information about the cisco-nsp
mailing list