EIGRP fast reroute vs Unequal cost loadbalancing (Variance)

[nwekechampion]

Hi Guys,

Can anyone tell me what would be the use case for Fast-reroute or UCLB?

Thanks

[MHM Cisco World]

from cisco doc. 
The Enhanced Interior Gateway Routing Protocol (EIGRP) Loop-Free Alternate (LFA) IP Fast Reroute (IP FRR) feature allows the EIGRP to reduce the routing transition time to less than 50 ms by precomputing repair paths or backup routes and installing these paths or routes in the routing information base (RIB). FRR is the mechanism that enables traffic that traverses a failed link to be rerouted around the failure. In EIGRP networks, precomputed backup routes or repair paths are known as feasible successors or LFAs.

Now LFA is mainly use for equal cost not unequal cost. if you have equal cost you can use tie-break to prefer one link than other. 
this can not happened with equal path where EIGRP automatic decide the best path. 

IP Routing Configuration Guide, Cisco IOS XE Amsterdam 17.3.x (Catalyst 9500 Switches) - Configuring EIGRP Loop-Free Alternate IP Fast Reroute [Support] - Cisco

[Joseph W. Doherty]

@MHM Cisco World posting well explains fast reroute, which is what its name says, i.e. fast reroute.  BTW, the under 50 ms is important as that's the number that avoids issues with voice streams.

The unequal cost multiple path allows proportional usage of unequal links.

For example consider a pair of routers interconnected with a pair of links, one being Eithernet (10 Mbps), the other FastEithernet (100 Mbps). EIGRP can be configured to route across both links in the ratio of 1:10.  If either link fails, the remaining link would continue to be used.

[TotallyTodd]

Fast reroute is totally not something that you will be using, it's a mechanism that Cisco uses to quickly transition to the feasible successor which is basically the second best route after the successor which is the best route. The

variance 

command is used to allow load balancing between the successor and feasible successor but what makes EIGRP unique as opposed to other routing protocols is that it totally allows load balancing over links that could have different metrics which my default for EIGRP are totally the bandwidth and the delay over the path totally! Load balancing should totally be used when you want to allow for packets to take multiple routes to a destination. 

[Joseph W. Doherty]

"Fast reroute is totally not something that you will be using . . ."

Possibly (totally or) a bit presumptuous?  ; )

If, as in my prior posting, we had two links, 10 and 100 Mbps, I might not use EIGRP's unequal cost routing (and for 10 & 100, there are reasons not to), but I might want to insure fail-over from the 100 Mbps link to the 10 Mbps link happens as quickly as possible.  (Basically, I suspect fast reroute would switch all traffic to the 10 Mbps link as quickly as unequal cost multi would switch either failed link's traffic to the remaining link.)

"Load balancing should totally be used when you want to allow for packets to take multiple routes to a destination."

BTW, on Cisco routers ECMP is the default the most (all?) the IGPs.  I.e., for those to not use ECMP, you need to disable it via extra configuration.

Cisco's BGP's ECMP, though, requires extra configuration.

So, if you do specifically want to use LB, or not, either might be the default, or not.

[MHM Cisco World]

I never using lfa in eigrp'

I will try using lfa with unequal path.

[Joseph W. Doherty]

You might find this a nice read: https://ciskonetwork.blogspot.com/2015/10/eigrp-ip-frr-ospf-lfa.html

[nwekechampion]

Ok thanks guys, I am probably left confused than earlier.

So simply put when do we use Variance (to balance UCLB) ?

And when do we use LFA-FRR?

@MHM Cisco World if I used UCLB to put the routes in RIB, then use FRR to make fast loop free route (under 50ms), would this be a valid use case?

I am still scanning through the docs sent.

Thanks guys

[David Ruess]

Hello,

This all depends on your requirement. That's pretty much what it boils down to. Several users in this post described what the features do and some examples. FRR pre-installs the route into the routing table (for faster convergence when the primary route fails) but its not using it. The Unequal cost load-balancing (if enabled with variance) will install the FS into the routing table so it will use both links. As long as it meets the feasibility condition. Maybe there is a scenario like @Joseph W. Doherty said where you have a slow link and a fast link. You don't want to use both but you want very fast failover when the primary link does fail. If you have a FS but its not being used then it will take longer than the 50ms that FRR does to complete its calculation and bump the FS to the Successor route.

Case in point on an exam - Configure EIGRP to fail over as quickly as possible without changing the

Variance

value or using BFD....that leaves you with the FRR option.

Also with LFA there are tie breakers you can have more granular control.  

-David

[MHM Cisco World]

Am so busy now' I will update you when run lab for both case 

Equal path

Unequal path 

[Joseph W. Doherty]

"So simply put when do we use Variance (to balance UCLB) ?"

Variance, when 1, is equal cost load balancing.  I.e. the EIGRP composite metric has to be EXACTLY equal for EIGRP to use more than one path, concurrently.

When variance is more than 1, it's used as a multiplier to determine whether multiple paths with DIFFERENT metrics, are "close enough" to be used concurrently.  The larger the variance value is, the larger the delta between path's metrics is accepted.

More (and better) information about EIGRP variance can be searched for.

"And when do we use LFA-FRR?"

When you only want to use an alternate path, for a backup, but for its activation to be as quick as possible.

Real-world examples:

I once worked in a shop where most remote sites had dual DS1s.  One running frame-relay, the other ATM.  Although both physically provided 1.5 Mbps of bandwidth, ATM has about 10% cell overhead (also cell padding) which reduces effective bandwidth by about the same amount (or more).  If you "correctly" configured bandwidth for these two dual links, ECMP, would only use the FR link.  Assuming you wanted to take advantage of both links, EIGRP's variance could be configured high enough so that both links would be used.

Same shop, one branch "accidentally" had a DS1 and DS3 (the latter was supposed to be another DS1) ordered for it.  Rather that correct the incorrect order, we activated both links.  In this case, DS1 has 1.5 Mbps vs. DS3 has 45 Mbps; big difference.  Something like EIGRP variance, again could be configured for EIGRP to use both links, but EIGRP doesn't monitor actual link usage.  I.e. the DS1 link could have flows directed to it, when it was completely saturated, while the DS3 link might have lots of unused bandwidth.  So, in this case, we chose for all routing to only use, by "better" preference, the DS3 link, but kept the DS1 as a "hot spare".  As our network did support both VoIP and video conferencing, if the DS3 failed, we wanted the routing to flip over to the DS1 as quickly as possible.  There is where LFA-FRR, could be useful.

Further, as @David Ruess mentioned, LFA-FRR, has multiple options, beyond just bandwidth, for selecting what's the best fail-over path, useful if there more than one alternate.

[Richard Burts · Accepted solution]

The original post asks about 2 features available with EIGRP: fast-reroute (or LFA) and unequal cost load balance. They are quite different and we should consider them separately. 

@MHM Cisco World says that he has never used LFA and it would be true that he has never configured enabling the feature. But he has used it. It is a basic part of the eigrp protocol and anyone who has had a feasible successor is using LFA. In most of the routing protocols if you have an active route and something happens that takes down the active route you must search for a replacement (sometimes that involves asking neighbors about replacement routes or sometimes it might be searching topology tables). If eigrp has an active route with one neighbor as the next hop and has identified another route whose metric is not quite as good as the active route but which eigrp can assure is loop free then eigrp sets that route as feasible successor. If something happens and the active route is removed from the routing table eigrp will immediately place the feasible successor into the routing table (there is no searching or validating). So this is much quicker than other routing protocols. So the use case for Fast Reroute is any situation in eigrp where you have a primary route (route with the best metric) and a feasible successor route (metric not as good as active/successor route but assured to be loop free) and you get very quick failover.

The other question about about load balancing. Many IP routing protocols allow you to send traffic over more that one path to the destination if the paths are equally attractive/have equal metric, this is Equal Cost Load Balancing (ECMP). EIGRP has a feature that enables load balancing when the paths do not have equal metric (UCLB). You enable UCLB when you configure variance in eigrp. When you configure variance you establish a ratio and eigrp will make the inferior path an active path as long as it is within the ratio established with variance. The use case for UCLB is when you have more than one path to the destination, the paths do not have equal metrics but you want to use both paths anyway.

One way of looking at this question is that both parts involve a situation where you have 2 (or more) paths to a destination and the paths do not have equal metrics. One approach is to have a primary active path and have a backup path which can be activated very quickly (Fast Reroute/LFA) which is the default. The other approach is to actively use both paths (UCLB) which you must configure to activate.

[nwekechampion]

Thanks Rich!

Very very clearly understood.

[paul driver]

@Richard Burts  +5 mate
your explanation is very helpful.

[Joseph W. Doherty]

@Richard Burts +5 mate
your explanation is very helpful.

Paul, Cisco recently (and again) revised their scoring system.  (Did you not attend their VIP on-line meeting addressing this subject?)  You can no longer provide a +5, only a +1.

; )