We are not a large organization by any means: we have an MPLS network with the HQ and 8 branch offices. We have a backup MPLS via another supplier.
In the past we had a 6513 at the core of our network and we used a combination of EIGRP with the variance command plus PBR to accomplish the following: we prefered one provider for certain mission critical traffic, unless that provider was down then all traffic went over the other link. As long as both links were up, non-mission critical traffc was shipped proportionately over the backup link and the remaining bandwidth of the prefered link. Ugly to look at, but simple to implement and worked like a charm.
Now, we bought expensive Nexus equipment to replace the 6513 and, well, EIGRP variance command is no longer part of the set.
I'm left with PfR on the two routers which is infinately more complex than what we need.
Every document I read about configuring OER/PfR is more essoteric than the one before it. It's not that there isn't good stuff out there, but getting to the heart of the matter "how do I configure a simple set up" is not to be had. The info I need to get to is burried under features, design, and marketing hype.
Is there anything out there that shows a very simple set up? I should have thought a simple access-list type definition for my prefered networks pointing them to provider A plus some way to ration out the remaining packets would not take so long to ferret out but here I am weeks later still reading and not making much progress.
To date I have GNS3 set up and PfR is "enabled and active" on the simulated links. I just need to translate my once-upon-a-time working ACL/PBR configuration into the PfR syntax. It's harder than it sounds.
Thanks for any links or examples.
I would recommend reading the PfR Doc Wiki: http://docwiki.cisco.com/wiki/PfR:Home
Let us know if you have further questions.
Steve Lyons - Cisco
Both these post links contain very helpful info and I've seen some other posts besides that do help but, still running into problem with this. Steve you touched on this in a conversation with our SE and myself on a conference call.
Here's the situation. I need...am MANDATED in fact, to prefer one provider for certain types of traffic, then load balance based on available bandwidth the remaining traffic.
Item one is easy: preferred link groups. But everything I read says that function is mutually exclusive with load balancing based on bandwidth. So what am I to do? Tell my employer the million dollar investment in Nexus was a mistake?
I hate to sound so frustrated but this PfR solution, while useful to large shops and ISP's, is convoluted to say the least, and overkill for a shop that has a fairly static amount of traffic over an two MPLS networks that simply aren't impacted by many (if any) external factors. Life was fine with EIGRP, variance, and a small number of PBR statements. While that wasn't exactly 'load balanced' perfectly, it was close enough to satisfy the requirements. It worked well for 5 years and because of how EIGRP worked, failure of any link was undetected by end users. Simple, clean, effective.
And here's an unrelated question that has plagued me throughout my studies of PfR: if I have a link that is suffering performance degredation do I really need both passive and active probes adding traffic to that link trying to determine if/when it's time to fall back to it? Sounds kind of like when I'm trying to fix a problem and my boss is constantly over my shoulder saying, "is it fixed yet, is it fixed yet?"! How exactly is this helping?
I really need variance back. without it, I'm not seeing anyway to meet my mandate.
whiel i do understand what you mean with this kind of surf-shoulder "is it fixed yet" i really do not feel like telling you that Nexus was a useless choice. Although i do not know your design and your infrastructure features, nexus and its NX-OS represent the only way for a more reliable infrastructure. Sounds clever to claim back EIGRP with PBR and variance but you will never get any reasonable comparison between load-sharing (EIGRP/PBR) and load-balancing(PfR). As Brian Dennis was telling many times in that seminar, people will get tired and sick soon to hear that STP is blocking links from working because of loop-free feature. Both your solutions work fine, load-sharing and load-balancing, but it is entirely ( i am afraid) to you chosing the way you wat to go. Should you be able t share some info more i can see it myself and try to help you more.
I'm not saying our method was better, I'm saying it was satisfying a requirement, that I now can find no way to satisfy.
We have two MPLS networks between our HQ and each branch. The following three types of traffic: voice, video, and citrix, which are all easily identifiable in a class map, MUST be sent over provider A unless provider A is down, then it may go over provider B. All the rest of the traffic (http, CIFS, etc) should be sent over either provicer, best effort is good enough (hence wise load sharing vs. true load balancing is not really a fine point with us).
One provider is 15 MB the other is 6 MB. Enter Nexus that won't let me put multiple paths in the EIGRP table unless they are equal becuase variance is now gone. So how do I get the traffic to go out a route that I can no longer put into the route table?
Well, we were told to install new routers that do PfR between them and Nexus.
But now I find PfR won't do both prefered links AND load balancing on bandwidth. Is there a way to make it prefer links only for one class and load balance the rest? Because that's not what I'm seeing out there, and that is what I must make it do.
At the end of the day considering that the administrative distance is superior you could deploy it in everywhere with no EIGRP disruption. After that you coudl simply switch the newtork working in EIGRP over OSPF and that's it. I did it for a 100+ sites 6000+ SP. It works fine and at that point is matter of cost of the link. Even on Nexus shouldn't be too hard to deploy it like ships in the night. I generally do not like to propose these changes but it does not make sense (as you have underlined) to buy two routers to do PfR.
i was thinking that you could even do some MPLS-TE. On Tunnel 1(crossing the 15Mbps link) you will give higher priority while on Tunnel 2 you can forward only determined types (EXP bits) of traffic or using it just as backup. Another advantage would be the encoding that MPLS does with OSPF. You would have your SuperBackbone and the complexity of a single OSPF domain.
Not to offend anybody, but among all the very skilled guys in NetPro, Giuseppe Larosa seems to me one of those who could address you on this. Email him!!
Hope this helps
i am not telling that OSPF is supporting unequal link load-sharing !! I am telling that it is much easier to deicde how much a link costs and therefore to address traffic where you need.
Huh…that’s interesting. We did EIGRP because OSPF (back in the day) didn’t support unequal link load sharing…are you saying it does? We don’t care which one we use (EIGRP or OSPF). Regarding routers and Pfr, we already drank the kool-aid, the routers are ours. One we purchased a while ago, the other was kind of a ‘deal’ dependent on two other purchase/tradeins as a way to solve the fact that we traded the 6513 for the 2 nexus and only then were we told it can’t do what we’d been doing…long story, and it doesn’t matter. I’ll contact the person you suggested. But I think we are going to be stuck with PfR at some level… thanks!
I am starting to work with Pfr and noticed that you have on GNS. What equipment did you use and what IOS are you using in GNS?
Thanks for the help