How many routes? (OSPF question)

[kjarri]

Hi everyone,

I have a question, how many routes

can ospf possibly carry without draining

the routers down from cpu cycles of processing

LSA's?

conditions:

5-10 routers,

single area,

all functioning with 7200 NPE-G1..

How many OSPF routes could be there

before OSPF do more damage than good?

[Harold Ritter]

The answer is a lot ;o) I have a feeling your next question will be how many routes is a lot, right?

We very often see customers running OSPF in topologies way more complex than this and carrying between 4000 and 6000 routes (without any kind of summarization) without too many issues.

In your case, you have a couple of factors playing for you.

- Powerful processors and lots of memory.

- Only ten routers in the area, which is rather on the low side.

- Probably very few internal routes (given the number of routers).

In short, I would worry too much about it if I were you.

Hope this helps,

[kjarri]

Every one of them has currently about 6000 of them,

locally but summarizes them when sending to his neighbors.. and its about 2000-3000 then..

The why im asking is, is it better solution to change

this into, OSPF handling loopbacks + links, and then

run BGP for those many routes.

because when i do #show proc cpu sort ,

OSPF router process and OSPF hello process, are always in the top 10 processes.

?

[Harold Ritter]

When you say 6000 routes, you mean 6000 external routes that you summarize when redistributing into OSPF, right? What is the origin of these routes?

The BGP approach would probably make sense.

The fact that OSPF processes are always part of the top 10 processes doesn't mean much if your CPU run at 1%. What is the current CPU utilization?

Thanks,

[kjarri]

Yeh, after summarization they are about 1000-2000 when going out of each router.

well the routers cpu are over 50% CPU usage constant,

very loaded routers.. depends when at the day, go higher at nights..

The fact is, is it wise to run OSPF for this kind

of work?

because of the cpu usage by link state calculations..

[Harold Ritter]

Q) The fact is, is it wise to run OSPF for this kind

of work?

A) I have to admit that it is a fair number of routes.

I believe BGP would be a good alternative. As you stated only the internal links and loopback addresses would be carried in OSPF.

Hope this helps,

[joyride_us]

Hi,

I have been confronted to more or less similar issue but with much smaller routers(2600).

I teted a lot of solutions, I see 3 alternatives if your routers are getting overloaded:

1. split your OSPF domain in many small OSPF processes and control the route redistribution with redistribute commands so that just the necessary routes are redistributed. Or even no link at all between the OSPF processes : just static routes with various costs to route only necessary traffic here and there.

2. change to EIGRP if you are very Cisco-oriented : I never tested it but Cisco promises that EIGRP has a very well optimized algorythm for route calculation that will take less resources from your routers..to be checked in real life though! But just the auto route-summarization is a very attractive feature if your IP addressing plan allows it! And teh max-bandwidth for signalling traffic is attractive too.

3. switch all to MPLS. Very efficient. No more routing protocols except on the edge routers. To tested too! I surely will next time!

Please keep us posted on your choices and results!

Cheers

Jef

[Harold Ritter]

3) MPLS still requires an IGP on all routers including the core routers. LDP propagates labels for IGP learnt prefixes (FEC) that are installed in the RIB.

Hope this helps,

[joyride_us]

Very interesting...I am currently evaluating MPLS to see if it unloads my routers a bit, especially on the memory utilization. Do you think it would?

Thank you

Jef

[Harold Ritter]

In a ISP core scenario, MPLS would substancially decrease the size of the routing table on the core routers if you don't run BGP on these routers. In this specific scenario, only the edge routers run BGP and carry the full Internet routing table. The core routers only need to carry the igp routes require to build the LSPs from any to any edge LSR.

Packet entering the core are labeled at the network edge and label switched on the core routers.

At a minimum you would need to carry all PEs loopback addresses in your IGP. You would also need to use the nexthop-self on all edge routers.

Hope this helps,