cancel
Showing results for 
Search instead for 
Did you mean: 
cancel
1599
Views
10
Helpful
5
Replies

EIGRP/OSPF neighbor scaling

billybjo1
Level 3
Level 3

I am about to design/implement a customer network that has about 200 router neighbors on a flat vlan (all Cisco). Apart from trying to ascertain if OSPF is a better choice over EIGRP, I can't find any sizing documents on CCO about router spec/cpu/memory considerations. I've seen various threads where its stated these have to be taken into account, but not actually how I go about sizing those choices. A few questions I need to resolve:

1) OSPF or EIGRP. I prefer the latter, but appreciate OSPF has some scaling boundary advantages. But can't see how those would help on a flat network. Some of the downstream networks already have EIGRP implemented. If I went for an OSPF area topology, surely I'd have to create a number of vlans to segment various groups of stub sites into areas, I couldn't do that on a flat vlan could I?

2) The vast majority of these sites are stub and have no downstream neighbors and only about 2-3 subnets, hence I could implement EIGRP Stub in those. But they would still have 200 neighbors each. There are 2 "core" sites each that have other networks downstream.

3) How do I size minimum router spec for the stub and core sites? Would 2800/2900/3800/3900 be feasible or am I looking at higher spec routers?

Many thanks.

1 Accepted Solution

Accepted Solutions

Hello Billy,

so my guess was correct that your WAN is a VPLS.

Scenario1 puts high pressure on devices.

Scenario1: EIGRP all devices even if remote site routers are stub they see each other in the shared vlan and they build EIGRP neighborship with all the others. traffic between two remotes sites flow directly without going via HQ.

when a remote site losts connection to one of its internal subnets it sends out an EIGRP query on the WAN shared link.

All other devices have to reply within the expected time or a stucked in active SIA will happen. All other remote sites will answer restating they are stub routers and Hub routers can:

answer with a negative answer if appropriatate EIGRP summarization is in place that includes the IP subnet object of the query.

or they can propagate the query to the internals of HQ.

The risk is flapping of EIGRP neighborships on the VPLS WAN when a fault happens.

OSPF: OSPF neighbor stare machine is more complex but the DR and BDR concept helps. Only DR and BDR, the two hub routers with appropriate OSPF priority, have to build full adjacencies to all the remote routers. each remote router will have two adjacencies with DR and BDR and simple neighborship with all the others.

In case of a local subnet failure the remote router sends an updated LSA to all DR addresses 224.0.0.6. DR will resend it to all SPF routers 224.0.0.5.

what is important in the case of OSPF is to distinguish between number of routers in the same area and the number of devices in the same IP subnet.

Having 200 routers in the same area is not a problem nowdays, having 200 routers in the same IP subnet is not the same.

The load for DR and BDR is consistent.

The only advantage of this scenario is direct communication between remote sites, but usually traffic patterns involve remote site to HQ if all servers are in HQ (mail server, file servers, and so on).

Scenario2:

here remote site to remote site communication can require to go through the HQ.

But it puts far less pressure on devices.

I would go for scenario2 putting remote devices in a few groups.

Using 802.1Q tags may be just a question of configuration, service provider can answer about this.

It may require configuration only in your side.

A flat network looks like attractive but when it comes to dynamic routing it is not a good thing with high number of devices.

Hope to help

Giuseppe

View solution in original post

5 Replies 5

Edison Ortiz
Hall of Fame
Hall of Fame

200 router neighbors on a flat vlan (all Cisco).

Define 'flat'

To me, it means all routers are on the same subnet

I can't find any sizing documents on CCO about router spec/cpu/memory considerations

Very hard to come up with a document where there is a ton of variables. Experience plays a factor here. If you haven't done it and you are not willing to take a risk, I recommend bringing someone in who has done it. If the business is willing to take a risk with someone who has not done it before, then this can be a very fruitful experience.

1) OSPF or EIGRP. I prefer the latter, but appreciate OSPF has some scaling boundary advantages. But can't see how those would help on a flat network. Some of the downstream networks already have EIGRP implemented. If I went for an OSPF area topology, surely I'd have to create a number of vlans to segment various groups of stub sites into areas, I couldn't do that on a flat vlan could I?

I'm confused again. You mentioned 'flat network' yet you mentioned downstream networks and areas. You can't do areas in a flat network. You can do areas in a hierarchical network. You also need to take into account how your customer is going to be able to manage their network after you leave. Are they proficient with OSPF or EIGRP or both? That must play a part in your design decision.

2) The vast majority of these sites are stub and have no downstream neighbors and only about 2-3 subnets, hence I could implement EIGRP Stub in those. But they would still have 200 neighbors each. There are 2 "core" sites each that have other networks downstream.

This is where OSPF shines as you can have each site on their own area and summarize at the site's ABR router toward HQ hence reducing your routing table dramatically as well as reducing routing updates.

3) How do I size minimum router spec for the stub and core sites? Would 2800/2900/3800/3900 be feasible or am I looking at higher spec routers?

Routing protocol shouldn't affect the decision on what hardware to purchase. The selected hardware should come from the type of WAN link, services (VoIP, for instance) and amount of throughput expected on such location. If you are looking for a T3 as the WAN link, you should consider no less than 3845 ISR.

Regards

Edison

Giuseppe Larosa
Hall of Fame
Hall of Fame

Hello Billy,

I share Edison's doubts.

>>  that has about 200 router neighbors on a flat vlan (all Cisco)

Do you mean your remote sites connect to the headquarters via a VPLS service so that they are all in the same IP subnet?

that could be your OSPF area 0 and remote sites routers could be ABR nodes with totally stub area for remote site IP subnets.

this roughly equates using EIGRP stub feature.

the real problem is the flat design for the WAN part that would cause a full mesh of EIGRP neighbors.

OSPF can save some resources thanks to DR and BDR concept so OSPF is advantaged in this crowded VPLS.

I strongly suggest to consider a way to put groups of remote sites in different IP subnets, if this is a VPLS service you should understand if you are free to use vlan based subinterfaces that use a different 802.1Q vlan-id.

A divide and conquer approach is to be preferred.

in other words you should emulate with 802.1Q tags what it is done in Frame Relay networks.

This can be justified by traffic patterns if most of traffic of each remote site is with HQ and hub and spoke is a good solution.

Hope to help

Giuseppe

paolo bevilacqua
Hall of Fame
Hall of Fame

Are you talking to your local cisco sales team ?

With 200+ sites they will be delighted to help in design and provide you with info not available elsewhere.

Beside that, I can only second Edison's considerations.

Where I say flat network, its a single subnet. Well to be more accurate its a single L2 

topology (single vlan .1q), there is no customer network overlaid as yet. It may well be VPLS, but all we see is an RJ45 presentation.

The 2 core sites are 1gb, some large remote sites are 100mb, and the vast majority are 10mb.

So basically the WAN facing side of each router attaches to this network.

Behind each remote site there are no further networks or routers other than the 2-3 local directly connected subnets.

Some of the 100mb sites and both of the core 1gb sites have other existing networks using EIGRP.

Customer doesn't really have any indepth knowledge of either OSPF or EIGRP. Its really just evolved over time given the ease of EIGRP to some extent.

Obviously OSPF requires more thought and design if deploying in a hierachical approach.

So which ever way we appporach this, EIGRP or OSPF, the 200 routers will have their WAN facing side in the same subnet.

We may be able to get the telco to provision the ability to break the WAN into further vlans which would enable us to provide a hierachical design.

So there seems two possible scenarios.

In scenario 1 with a single subnet/vlan, all 200 routers will see each other directly. Have you any opinion if using EIGRP in that case (95% stub sites) would it be reasonably stable, even after a major WAN event causing convergence. If using OSPF (single area 0) then the same question.

In scenario 2 with multiple subnets/vlans, we could group the remote sites into 5 logical areas the way I see it. I'd have to examine their current subnet configurations to see if there are some natural summarisation boundaries, but we may be able to redesign that if not. If the customer is reluctant to change remote subnets, we'd just have to maybe accept there won't be a great deal of summarisation going on. If none was possible we'd really only be talking about 700-800 routes max.

I appreciate scenario 2 may well be deemed the better approach, but I'd like to explore the pros/cons of scenario 1 before ruling it out. Thats really where I started to consider router CPU/Memory requirements which I understand are not an exact science and other services should be taken into account. Often routers are selected on throughput/interface speeds, but I was hoping to define a more "scientific" approach to ensuring a minimum spec is at least capable of 200 neighbors/full mesh should we still end up with scenario 1.

Hello Billy,

so my guess was correct that your WAN is a VPLS.

Scenario1 puts high pressure on devices.

Scenario1: EIGRP all devices even if remote site routers are stub they see each other in the shared vlan and they build EIGRP neighborship with all the others. traffic between two remotes sites flow directly without going via HQ.

when a remote site losts connection to one of its internal subnets it sends out an EIGRP query on the WAN shared link.

All other devices have to reply within the expected time or a stucked in active SIA will happen. All other remote sites will answer restating they are stub routers and Hub routers can:

answer with a negative answer if appropriatate EIGRP summarization is in place that includes the IP subnet object of the query.

or they can propagate the query to the internals of HQ.

The risk is flapping of EIGRP neighborships on the VPLS WAN when a fault happens.

OSPF: OSPF neighbor stare machine is more complex but the DR and BDR concept helps. Only DR and BDR, the two hub routers with appropriate OSPF priority, have to build full adjacencies to all the remote routers. each remote router will have two adjacencies with DR and BDR and simple neighborship with all the others.

In case of a local subnet failure the remote router sends an updated LSA to all DR addresses 224.0.0.6. DR will resend it to all SPF routers 224.0.0.5.

what is important in the case of OSPF is to distinguish between number of routers in the same area and the number of devices in the same IP subnet.

Having 200 routers in the same area is not a problem nowdays, having 200 routers in the same IP subnet is not the same.

The load for DR and BDR is consistent.

The only advantage of this scenario is direct communication between remote sites, but usually traffic patterns involve remote site to HQ if all servers are in HQ (mail server, file servers, and so on).

Scenario2:

here remote site to remote site communication can require to go through the HQ.

But it puts far less pressure on devices.

I would go for scenario2 putting remote devices in a few groups.

Using 802.1Q tags may be just a question of configuration, service provider can answer about this.

It may require configuration only in your side.

A flat network looks like attractive but when it comes to dynamic routing it is not a good thing with high number of devices.

Hope to help

Giuseppe

Getting Started

Find answers to your questions by entering keywords or phrases in the Search bar above. New here? Use these resources to familiarize yourself with the community:

Innovations in Cisco Full Stack Observability - A new webinar from Cisco