EIGRP or OSPF for IGP?

[Cooky]

We're debating on which protocol to use as the IGP for our remote sites.
The remote sites are separated by BGP, so whatever IGP they run will be independent from each other.

EIGRP seems to be easier to configure, but OSPF provides interoperability w/ non-Cisco vendors.

What's everyone running?
Anyone has any recommendations?

 

[spidey07]

EIGRP has advantages over OSPF - faster convergence, no flooding, easier

I used to be a OSPF only kind of guy and it's still my specialty, but I've gone over to the EIGRP fence after using it in some very large networks. OSPF still has advantages when you start interacting with other routing protocols IMHO because of the different types of external route and want to do serious summarization/route optimization.

If you're going to be exchanging routing with any non cisco device then of course OSPF is your choice. Just make sure you have people on staff that are good at it, it's VERY powerful and with that power comes some complexity and ability to really screw things up if you don't know what you're doing.

If you're only going to be redistributing on a couple of devices then you could run EIGRP internally and then do mutual redistribution with OSPF on just those routers.

 

[Cooky]

We're a Cisco shop, so the only time we'll need to run OSPF is if we ever switch to Juniper.

In terms of interacting w/ other protocols...we just need to redistribute the IGP into BGP at remote sites.

Does it sound like we're suited for EIGRP?

Half of us voted for EIGRP, and the other half voted for OSPF, so we're now in a deadlock.

 

[mcmilljb]

Originally posted by: Cooky
We're a Cisco shop, so the only time we'll need to run OSPF is if we ever switch to Juniper.

In terms of interacting w/ other protocols...we just need to redistribute the IGP into BGP at remote sites.

Does it sound like we're suited for EIGRP?

Half of us voted for EIGRP, and the other half voted for OSPF, so we're now in a deadlock.

Why do they want to run OSPF?
Do you usually use EIGRP?
Do you have enough people who understand and/or have have implemented OSPF?

 

[spidey07]

When you do a OSPF vs. EIGRP it's like brining up religion or politics at the dinner table. It's not going to be pretty. People are generally entrenched in one or the other.

 

[drebo]

Piss off everyone and use IS-IS!

 

[spidey07]

Originally posted by: drebo
Piss off everyone and use IS-IS!

eww. Just saying it makes me puke a little in my mouth.

 

[yoda291]

you should flip a coin...

cuz really, you'll just be getting all sorts of divergent information from both sides on this fence.
"It's proprietary!"
"It's slower!"
"It's chattier!"

 

[cmetz]

Cooky, why not just use RIP? I mean, if you're going to make a bad choice of routing protocol and pick something that's dead for good reason, why go half way? RIP is the clear choice. And none of this pansy new-fangled RIPv2 stuff, either!

Even Cisco is now saying EIGRP is dead. Good riddance.

OSPF is a good protocol. It's not perfect, but it really is a good protocol. John Moy is a very, very smart guy, and they had a lot of other very smart people in the OSPF working group helping with the protocol design and specification. A lot of the historical aversion to OSPF was that Cisco's OSPF implementation was poorly written from the beginning, and Cisco considered that a feature rather than fixing all their bugs. Enterprise customers got forced into EIGRP if they wanted something they could rely on, and carrier customers got forced into IS-IS(*). Then Juniper came along, with a nearly bug-free OSPF implementation, Cisco did an "oh <explitive deleted>," and decided to actually start fixing their OSPF bug backlog. Now Cisco's OSPF works well enough for production use. EIGRP is dead, it's a flawed protocol at a theoretical level, it's proprietary, and it hasn't been getting updated with the times. If it's Cisco proprietary and Cisco is telling you to migrate away, take that clue.

(* - a lot of carrier folks will insist that IS-IS is really a better protocol than OSPF for a variety of contrived reasons. OSPF and IS-IS vary in ways that come down to potato vs. potatoe, but nonetheless there are folks who insist that their way is the one true and correct way. In practice, most real ISPs use IS-IS as an ISP and are not about to change, so if you were building a big ISP you'd be smart to choose what every other big ISP is using just because every other big ISP is using it. But, yes, the ISO-isms are sickening.)

OSPF's convergence can be improved through the use of timer values other than the defaults - IMO the default values are really silly long.

 

[Cooky]

cmetz - I believe I've heard similar comments from you about EIGRP being dead before.
Could you please jog my memory exactly why that is, and how reliable your source is on Cisco telling its customers to migrate from EIGRP to something else?
We have never heard that from our SE.

We run EIGRP at our headquarters and a few other remote sites, and we haven't really had any issues, except when we redistribute EIGRP into BGP, the requirement to define the metric is a bit annoying.

I believe ISP's run OSPF or IS-IS because that's a requirement to run enhanced MPLS. (forgot exactly in what regard though)

 

[spidey07]

Cmetz,

I think it all boils down to the common networking concept - "it depends". I love IS-IS from a conceptual design (the hierarchy), but not in practice...you had better have good designers similar to OSPF and competent staff to design and support it. That probably stems from my exposure being OSPF. Large ISPs have large/competent staff.

I'm not saying one is better than the other, but you need true experts to do it right. If OP needs a real IGP and BGP at the edge then he shouldn't be looking at EIGRP. It all depends on the details. Networking folks are bullheaded anyway.

Cooky - ISPs run OSPF/ISIS because they don't have all cisco gear and don't want proprietary shit...I don't blame them.

 

[Cooky]

Yes, not having all Cisco gear is one of the main reasons.
What I was referring to was a link state protocol is a requirement to run MPLS traffic engineering. (just looked it up)

 

[spidey07]

Cooky, post what you are really trying to do. "it depends"

 

[drebo]

I wasn't really being serious about IS-IS. I think it's really neat from a conceptual point of view, but for a two-site location, I don't think it'd be appropriate.

If the organization is small and doesn't require the specialized network types OSPF provides I'd say it's pointless to go through the pain of tuning OSPF and just use the easier-to-set-up-and-understand EIGRP.

 

[Pheran]

Generally I recommend using OSPF as an IGP unless you have special requirements, since EIGRP is just too proprietary. Even if you have all Cisco gear, it's no guarantee of compatibility. For example, Cisco ASAs did not support EIGRP at all until very recently, and there are probably still other Cisco products that are OSPF-only, especially if you have some older ones.

 

[Cooky]

We have about 80 remote sites, each doing static routing, and redistributing those static routes into BGP.
We'd like to migrate from static to a dynamic IGP.

As of now, we're leaning towards EIGRP, since we already run it in some of the sites, plus we exclusively use Cisco gear. (except SSL VPN, we went w/ Juniper)

Thanks for everyone's input on this.

*Edit*
cmetz - could you let us know where you heard Cisco recommends its customers to migrate away from EIGRP?

 

[cmetz]

Cooky, folks I know who are real engineers at Cisco have been telling me this for years. Look at what protocols they are adding new features to aggressively and which ones are just barely keeping up. If Cisco intended EIGRP to win, they would have taken the approach of making it a publicly available standard that they just happen to have the best implementation of. They know full well that a lot of important and smart customers won't use proprietary protocols, period. They've done it before and would do it again, if that's the technology they really want to win, they'll document it and make it a public standard, just make sure that they are way ahead of everybody else competitively. (In this respect I give Cisco a lot of credit; a *lot* of good public standards have come out of Cisco, and it's a reasonable price to extract that they get a competitive head start.)

The Cisco SEs I've worked with have quite consistently told me that I should be using the most proprietary and expensive solutions Cisco makes. So they're professional liars and I treat their advice accordingly. I'm sure that there are good SE teams at Cisco who treat things as a win-win opportunity, I just have always gotten the other kind. My SEs tell me to use EIGRP wherever possible, while the folks I know in engineering think that's good for a laugh.