Largest single-area OSPF network

[badalam_nt]

Do you have any idea what could be the size - in number of routers - of some big real single-area OSPF networks, which are working fine ?

I saw mentioned on some other websites a single-area OSPF network composed of 1000 (one thousand) routers working fine.

Side-questions, not related to single area OSPF or multi-area OSPF but for any type of routing protocol:

- what is the approx. size of the largest AS (autonomous system) network in terms of number of routers ?

- what is the approx. size of the largest area running a same IGP inside an AS (autonomous system) network in terms of number of routers ?

[Mohamed Sobair]

Hi,

There are a lot of factors in order to determine how many routers per Area and how many interfaces belongs to an area.

I would recommend you go through the following link to determine the Type Of Networks involved, How much routes are injected from the ASBR and the totall memeory required for every router.

From OSPF Design Guide:

http://www.cisco.com/en/US/tech/tk365/technologies_white_paper09186a0080094e9e.shtml

HTH

Mohamed

[badalam_nt]

My question is for what currently exists in real networks.

Don't need it for a design, so the answer I expect should be straight, providing a figure and eventually where seen this (if not confidential this latter part).

I was actually surprised to hear that such a large (1000 routers) could be in a single-area OSPF network and then all questions popped-up to my mind.

[Giuseppe Larosa]

Hello Petru,

old practical rule was no more then 60 routers in area 0.

However, links are more reliable and less error prone and there is less probability to have them full.

The router cpu themselves are much more powerful.

OSPF partial recalculation can help.

I've worked for some years in a project for a mobile operator and their OSPF area 0 included 120 routers: C7500, GSR and MSFC2 on catalyst switches.

It was not a problem.

However, I don't expect to see real networks with 1000 routers in area 0 even if it can be feasible.

About scaling the networks on big providers they use to partition the network in multiple sub-ASes with BGP confederation and each mini-AS can have its own IGP.

Hope to help

Giuseppe

[missiongeek]

ASFAIK its 50 Routers. But don't quote me on this.

[badalam_nt]

What Jeff Doyle advertised on a blog is that a single-area OSPF is almost always sufficient enough, mentioning a case with 1000 routers working fine.

So my understanding then was that actually one never need to deploy OSPF multi-area, always for OSPF practically one area only meets the requirements and works fine. (I can't imagine how a network can have 1000 routers...it is far too much, isn't it?)

[Giuseppe Larosa]

Hello Petru,

OSPF multi-area allows for fine route control:

you can use area filter-list command to decide which inter-area to propagate.

Inside a single area OSPF domain you cannot do that

In addition you should consider the fact that SPF is not executed at all in other areas if proper route summarization is performed with area range command:

if a component route fails but at least another component subnet is alive the summary route is alive and no change is detected outside the area.

Hope to help

Giuseppe

[badalam_nt]

Thanks Giuseppe.

I will soon jump into my next CCNA topic: OSPF (just finished RIP, though I still have some doubts about RIP, such as hold-down process, poison reverse - will post 2 questions for that) and before doing that I started first to read light general stuff about OSPF.

This is the excerpt from Jeff Doyle's blog:

So how big can a single OSPF area get? (...): I was once brought in by a network operator to assess their single-area OSPF network that contained just over 1,000 routers. Although they had not experienced any problems, they were planning to make some additions and were worried that the changes would be enough to tip the whole thing over. I did a careful analysis and could find no performance or memory problems; the customer made the planned changes and the network is still humming along in a single area."

[Joseph W. Doherty]

With due regard to Jeff Doyle, 1,000 does seem a bit excessive.

The issue with a large number of routers within a single area is SPF calculation. Or course, it's not just the number of routers but what the topology looks like. I.e. how complex is the SPF tree each router must work with.

The SPF issue can quickly become a problem if there is any link instability causing routers to keep recomputing SPF. I've seen over a hundred routers run well when everything is stable and seen twenty meltdown when a link was flapping.

Also when working with OSPF, not every vendor's implementation is as robust. The OSPFv2 "standard" leaves much up to the implementation. It's interesting to read some of Cisco's published internal enhancements they made to OSPF over the years. Some examples include incremental SPF calculation and LSA pacing.

Recently I was somewhat surprised to discover 3750s were much harder stressed by routing protocol calculation than 2801s. Normally, I think the L3 switch offers so much more performance than an ISR. Yet, upon reflection, the L3 switch obtains its performance from dedicated hardware designed to forward packets, not, perhaps, for much control plane processing. I make this point because how large you might build an OSPF area will likely also depend on the hardware's performance and "modern" MLS devices might not provide control plane performance as well as "older" pure routers.

In summary, although one network may have worked fine with 1,000 routers within an OSPF area, I wouldn't agree this means we might never need to use multiple OSPF areas.

One purpose for OSPF areas is to insure stability, but others exist. Since as other posters have noted, your OSPF route filtering is controlled at area boundaries, area design can, for example, minimize LSA bandwidth utilization, which might be especially important for "slow" WAN links.

[saikonetworks]

The number of *routers* in a OSPF area can be very high, likely into the thousands, as long as the IGP is used minimally on connected links and loopbacks, in a single area.  in a typical carrier network, most of the customer routing is handled through BGP with the IGP regulated to infrastructure, and single areas are preffered in situations where you're using MPLS TE.