dslreports logo
 
    All Forums Hot Topics Gallery
spc
uniqs
56
wvvawireless
join:2005-01-15
Rich Creek, VA

wvvawireless to unwired9

Member

to unwired9

Re: [Tech Ops] MPLS

We are also looking at going to pppoe and debating central vs pop termination and using mpls.

I also like the idea of the central termination but have a question.

We have an ospf network with redundant back hauls on each end of the network. Currently half the traffic goes to one and the other half to the other. If anything fails all traffic moves to whats available. If I use mpls tunnels and a concentrator back to one side how can I have it switch to the other upon a failure?

TomS_
Git-r-done
MVM
join:2002-07-19
London, UK

TomS_

MVM

MPLS typically switches over a single path when using LDP as your label protocol, and that usually follows the same path that your IGP thinks is the best way to route traffic around your network. So the short answer is yes, MPLS tunnels will automatically "switch" around failed links.

If you want to get really tricky you need to look at traffic engineering with things like RSVP. But even I havent really played with that.

If you dont mind statically assigning IPs to your customers (which is almost a requirement these days anyway with always on broadband) then you can still use localised termination, with the IP address assigned through RADIUS when the user authenticates. This way you still get the most efficient use of your IP space by not having to assign pools to each tower or BRAS, since your IGP will make sure the network knows how to route packets to each customer.
unwired9
Premium Member
join:2008-04-08
Algoma, WI

unwired9

Premium Member

I'm not sure why it was suggested other than it's a really good way to do things. My original question to them was how to handle the ip pool within radius and have the nas servers at each pop. What I am currently testing is I have added one of my public /24 into the ospf networks - So I have a central pool of addresses so I don't necessarily have to assign statics and I shouldn't need to assign blocks to each pop. The radius server assigns and address from the pool and then ospf takes over and adds in a route for the address on that corresponding router. The one issue with this config is going to be the routing table size but I feel I'm a few thousand subscribers away from that being a problem.

TomS_
Git-r-done
MVM
join:2002-07-19
London, UK

TomS_

MVM

said by unwired9:

The one issue with this config is going to be the routing table size but I feel I'm a few thousand subscribers away from that being a problem.

What sort of routers are you using? With a software based router (i.e. no hardware forwarding), routing table size should be limited only by how much RAM the box has.

If its a hardware based router, chances are it will support perhaps hundreds of thousands of FIB entries anyway.

There is also a RIB which is stored in RAM, and is used to populate the FIB. The RIB is usually much bigger than the FIB, since the RIB contains all copies of routes received, while the FIB contains only the ones needed to forward packets.

FIB = Forwarding Information Base
RIB = Routing Information Base

As a side note, the current Internet routing table at I think somewhere around 440,000 routes only consumes maybe 50MB of RAM on a Cisco router.

Inssomniak
The Glitch
Premium Member
join:2005-04-06
Cayuga, ON

Inssomniak to wvvawireless

Premium Member

to wvvawireless
said by wvvawireless:

We have an ospf network with redundant back hauls on each end of the network. Currently half the traffic goes to one and the other half to the other. If anything fails all traffic moves to whats available. If I use mpls tunnels and a concentrator back to one side how can I have it switch to the other upon a failure?

Yes just to add. It follows LDP. I have several redundant paths thru my network and it switches over just fine.
unwired9
Premium Member
join:2008-04-08
Algoma, WI

unwired9 to TomS_

Premium Member

to TomS_
I have mostly RB50g or 493ah at each pop - Imagestream at the core. I believe at most I have seen my routers spike at 15% cpu and that is with Butch's QOS script running. I will have to relocate Butch's scripts moving forward but was struggling to see just where the benefit in mpls would come vs what I already have in place.
bairdmj
join:2009-12-30

bairdmj to unwired9

Member

to unwired9
If you centrally assign IP addresses via RADIUS (from one common, central pool), you will inject one host route (a /32) into your IGP's (OSPF) routing table for every client who connects. Furthermore, every time one of your clients disconnects/reconnects their PPPoE connection, an update to your routing table will occur. Each device that participates in OSPF on your network will will receive the "update" every time a PPPoE connection state changes.

If you choose to route a block of IPs to each PPPoE concentrator, it will be a bit more difficult to manage your IPs efficiently. But, you will be able to summarize the route for the block of IPs. This means that instead of having 253 /32 host-routes for each client in a /24, the router/concentrator would only announce one summary route (the /24) to the rest of your network. In addition, because you are summarizing the route, your routing table won't update each time a client connects/disconnects.

Inssomniak
The Glitch
Premium Member
join:2005-04-06
Cayuga, ON

Inssomniak

Premium Member

said by bairdmj:

If you centrally assign IP addresses via RADIUS (from one common, central pool), you will inject one host route (a /32) into your IGP's (OSPF) routing table for every client who connects. Furthermore, every time one of your clients disconnects/reconnects their PPPoE connection, an update to your routing table will occur. Each device that participates in OSPF on your network will will receive the "update" every time a PPPoE connection state changes.

You can still filter this out no problem. I have a centrally assigned IP address setup from RADIUS and just block it in an "ospf-out" chain so all my other routers cant see it, they have no need to see it.

TomS_
Git-r-done
MVM
join:2002-07-19
London, UK

TomS_

MVM

Yeah. Most likely you can have the router doing the termination announce a summary route in place of all of the host routes.

If youre terminating locally at each tower or in a non-centralised way, naturally you have little choice.