The MPLS-OPS Archive

Cell Relay Retreat>MPLS-OPS Archive>month:2001-Mar> msg00179



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

Re: MPLS/VPN question

  • From: Robert Raszuk <raszuk@cisco.com>
  • Date: Mon, 26 Mar 2001 23:16:44 -0800
  • CC: saqibj@margallacomm.com, mpls-ops@mplsrc.com
  • Organization: Signature: http://www.employees.org/~raszuk/sig/
  • Resent-Date: Tue, 27 Mar 2001 03:45:53 -0500
  • To: Alexander Marhold <alexander@marhold.at>

Hi Alexander,

> On the other hand MPLS/VPN as L3-peering-technology has a much higher
> convergence-time than a direct FR-circuit.

Hmmm interesting. I would say that this is never much higher and also
can be sometimes even better. 

The reason why this is not much higher is that there is quite a few
networks running IGPs like EIGRP or OSPF where a convergence is equally
proportinal to: 

* the number of routers in a flat area + 
* configuration for LSA flooding optimization + 
* CPU power of your routers. 

In MPLS-VPN the convergence is quite deterministic as in most cases it
depends on the following constants:

* configuration of min adv interval on the PE-CE EBGP case (if you use
IGP on PE-CE you don't have this delay), 
* confiuguration of the import bgp scanner on PE, 
* CPU power of the edge PEs and reflectors used (which in most cases are
higher then the low end units connected by F/R PVCs) of the customers. 

The achivable goal for mpls-vpn convergence is to make it equal to IGP
convergence time in the flat area of the same size. Adding all elements
together it should not exceed 20-40 sec.

Why/When it can be better ? When you have a large set of sites connected
in a close to full mesh VPN via L2 media and most of those sites use low
end routers. Those would convergence wise benefit well from MPLS-VPN
offer by any of the provider.

R.

> Alexander Marhold wrote:
> 
> Hello !
> 
> My 2 cents on that as this is a kind of question I often see and get.
> 
> 1.) the main task of the PE (Ingress/Egress LSR) is to allow the MPLS/VPN
> based forwarding on the control plane and on the data plane.
> Regarding QOS the PE as service provider device runs the DIFFSERV model.
> Typically outside the PE sits an CE which is the USER-Interface.
> 2.) Things like selection into different VPNs. Address Translation,Protocol
> Conversion,protocol specific actions and functions....
> are IMHO  CE functions and not  PE ones.
> 
> So regarding the actual question: if you need different VPNs for the
> traffic, than separate it on the CE and have more than one (logical) CE-PE
> connection going into different VPNs.
> It is also not recommendable for any Service Provider to do
> customer-specific things on the PE.
> 
> Generally spoken, does it make sense to convert an existing FR-network to
> MPLS/VPN ?
> As always, it depends
> Despite the wise rule "never change a running system", there maybe some
> reasons
> COST, COST,.....
> Main advantage of MPLS/VPN is an integrated FULL-MESH whereas FR is a
> circuit point-to-point technology. So if you have to any-to-any connect 100
> sites, MPLS/VPN gives you distinct advantage in that case.
> On the other hand MPLS/VPN as L3-peering-technology has a much higher
> convergence-time than a direct FR-circuit.
> ....
> 
> with best regards
> 
> Alexander Marhold
> Senior Consultant
> PRO IN
> 
> -----Original Message-----
> From: Saqib Jang [mailto:saqibj@margallacomm.com]
> Sent: Tuesday, March 27, 2001 5:50 AM
> To: raszuk@cisco.com
> Cc: mpls-ops@mplsrc.com
> Subject: RE: MPLS/VPN question
> 
> I think my choice of VoIP (due to the QoS issues surrounding it)
> as an example may have confused the issue. Assuming a number of sites have
> set-up a FR based overlay network to forward a specific type of traffic
> such as HTTP, FTP, or streaming traffic (setting aside the QoS implications
> for now). Does it make sense for them to look to get rid of the overlay
> network and use their existing last mile links thats used for
> all their data traffic in conjunction with an MPLS VPN that supports only
> (HTTP, ..) the specific type of traffic supported by their
> existing FR network. Or will MPLS VPNs not work for such a scenario?
> 
> Saqib
> 
> -----Original Message-----
> From: Robert Raszuk [mailto:raszuk@cisco.com]
> Sent: Monday, March 26, 2001 3:41 PM
> To: saqibj@margallacomm.com
> Cc: mpls-ops@mplsrc.com
> Subject: Re: MPLS/VPN question
> 
> Saqib,
> 
> I don't think that the approach of application specific VPNs even makes
> sense. L2 or L3 VPNs provide you mainly with connectivity (routing
> information distribution).
> 
> For application level scheduling guarantees you should be using the
> right QoS tools available both in each hop as well on ingress. That way
> you can classify very important traffic (VoIP for example) to be taking
> the dedicated LLQ resources at each hop. I don't know how building VPNs
> (routing information disgtribution) may replace basic qos engineering in
> your network.
> 
> Also the existance or not of NMPLS-BGP VPNs is orthogonal to the
> possibiliy of using more complex QoS tools like diffserv aware TE which
> can be a perfect complement to the VPNs itself, but not a reason for it.
> 
> R.
> 
> > Saqib Jang wrote:
> >
> > As I understand it, MPLS/BGP VPNs (per RFC2547) enable site-to-site
> > VPNs at for all traffic flowing between sites in the VPN. Can MPLS
> > VPNs handle the requirement of creation of VPNs among sites for
> > specific types of traffic (say e.g. VoIP traffic)? Would this require
> > a VoIP swtich (other type of allication-aware switches) to be an MPLS edge
> > router or can a "one size fits all" approach work here (i.e. can standard
> > MPLS-aware IP routers create application-specific MPLS VPNs). I'm looking
> > for technical rationale (i.e. not company/product positioning) here.
> >
> > TIA,
> > Saqib
> >
> > Saqib Jang
> > Margalla Communications, Inc.
> > 3301 El Camino Real, Suite 220
> > Atherton, CA 94027
> > Ph: 650 298 8462
> > Fax: 650 851 1613
> >
> > -------
> > 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