EIGRP: Delay & Variance

EIGRP variance and delay is the one subject I am having trouble with in ICND2. Can someone elaborate on delay and variance in easy to understand terms?

I am using Boson to prepare and even their explanations confuse me.

If someone can explain what the delay and variance does in the simplest and shortest way possible that would help tremendously.

Thanks.

Find more posts tagged with

Comments

DCD

Delay is the amount of time a packet takes between point A and point B. Variance is if you have more than one path to a destination and you split the packets between the various paths.

RouteMyPacket

Let's be more specific here. Variance is for what? Unequal Cost Load Balancing right?

When we have multiple (two) paths to a destination that are unequal cost (have different metrics), we can use the "variance" parameter to determine which route is used by the router.

This is a good demonstration of Variance and EIGRP Load Balancing, take a look

How Does Unequal Cost Path Load Balancing (Variance) Work in IGRP and EIGRP? - Cisco

Delay on the other hand is a metric used by EIGRP and we can manipulate the delay on a particular interface to either prfer that route or ensure it is say a secondary option etc.

Say we have two interfaces participating in EIGRP 100, Tun1 and Tun2...we want to ensure traffic traverses Tun1 as primary and Tun2 as a backup, we can configure under the Tun2 interface something like "delay 100000"

How do you determine delay? Look at the interfaces

show interface Tun1
show interface Tun2

Then ensure delay is higher on Tun2, boom..done!

theodoxa

Delay along with Bandwidth is one off the two main variables (you could also use load and/or reliability, but this isn't generally a good idea) in the calculation of EIGRP's metric. Basically, it uses the formula (assuming K Values are default)256 x ((10 Gbps / Least Bandwidth) + Sum of Interface Delay in 10 microsecond increments)Cisco assigns interfaces a default delay depending on the type of interface and speed, which I presume is supposed to be an estimation of the delay experienced on that type of interface. You can change this by configuring the delay under interface configuration mode, allowing you to tune the metric (this is something you'll run into at the CCNP and above level).

https://learningnetwork.cisco.com/thread/6116

As for variance, EIGRP supports Unequal cost load balancing, which means traffic can be load balanced over multiple routes whose metrics are NOT the same. The variance is simply a multiplier. If route #2 has a metric of less than or equal to route #1 (best route)'s metric multiplied by the variance AND route #2 passes the feasibility condition, it will be used for load balancing.

EIGRP_min.png

White Wizard

Thank you everyone, great and simple explanations.

mikeybinec

So the optimal path to the loopback on the right side router would be bandwidth 1024 and that would be
the route you would see in the routing table.. Increasing the variance would include the lower bandwidth
route also in the route table thus called unequal load balancing

eigrp bandwidth.jpg

theodoxa

256 ((10000000000 / 64000) + 2000 + 500) = 40,640,000
256 ((10000000000 / 1024000) + 2000 + 500) = 3,139,840

40,640,000 / 3,139,840 = 12.94

A variance of 13 should work

Trying to figure out the difference (PT vs. My Calculations), I realized the Loopback interface has its own Bandwidth (8 Gbps) and Delay (5000 usec) that also have to be taken to account.

Router#sh ip eigrp top

IP-EIGRP Topology Table for AS 10

Codes: P - Passive, A - Active, U - Update, Q - Query, R - Reply,
r - Reply status

P 10.0.0.4/30, 1 successors, FD is 3011840
.........via Connected, Serial0/1/0
P 10.0.0.0/30, 1 successors, FD is 40512000
.........via Connected, Serial0/0/0
P 192.168.200.1/32, 1 successors, FD is 3139840
.........via 10.0.0.6 (3139840/128256), Serial0/1/0
.........via 10.0.0.2 (40640000/128256), Serial0/0/0

Router#sh ip route eigrp

192.168.200.0/32 is subnetted, 1 subnets

D 192.168.200.1 [90/3139840] via 10.0.0.6, 00:00:05, Serial0/1/0
................[90/40640000] via 10.0.0.2, 00:00:04, Serial0/0/0

mikeybinec

Hey Theo--just a quick question that is somewhat related to the OPs question.

Sticking with my diagram I posted a few messages ago--when I run the sh ip ro command I see load balancing to the loopback interface. Now, if I change the bandwidth on the left router, say bandwidth 64, why doesn't the right side peer adjust to the change in bandwidth--Kinda like one side is half duplex and the other is full duplex. If I lower the BW on the left side router, that path/route will be dropped from the sh ip ro table. But the right side router is still at it's configured BW..

Do I make sense?

Thanks

squirrelonfire

1) I actually wrote about my understanding on these 2 particular features in my blog

%DUAL-3-SIA: Route network mask stuck-in-active state in IP-EIGRP AS. Cleaning up: EIGRP

Just think about it like this, in order to load balance 2 links, they have to have the same Feasible Distance right? because if they aren't, the higher Feasible Distance path will be placed in the Topology table as the Feasible Successor (yes that path must also meet the Feasible Condition too but assume it does for simplicity sake).

But what if you want to use a path that DOES NOT have the same Feasible Distance as the Successor, in load balancing? How can you do that? This is why you use Variance.

Variance is to make a Feasible Route become a Successor route so that you can load balance it. Even though you know this Feasible Route will never qualify to become a Successor route. But Variance will make it become so.
So you calculate the variance by diving the FD of the Feasible Successor by the Feasible Distance of the Successor.
This variance is now effectively increases the minimum Feasible Distance of the Successor route so that the Feasible Successor's Feasible Distance will appear as if it is equal to this number. And because now it seems like it is equal, EIGRP will put this Feasible Successor into the Routing table to load balance it.
Here is some math,
Successor's Feasible D is say 10
Feasible Successor's FD is say 230
Variance is 230/10 = 23

Now the route thinks the FD of the successor is what? 23*10 = 230.

Now the FD of the Feasible Successor looks like it is actually the same as the FD of the Successor 230 right?

Now the router will use the Feasible Successor path in its routing table, which will effectively use it for Load Balancing.

2) Delay is just 1 way of influencing EIGRP to prefer 1 path to another as the main routing path.

Say you have 2 routers R1 and R2 both have 2 paths connect directly to it.
1st path is Serial to Serial
2nd path is Gigabit to Gigabit

No EIGRP will not load balance this because the Serial to Serial path will be chosen as Feasible Successor only.

But hey what if you want to use the Serial to Serial as the routing path and the Gigabit just as back up? You just increase Delay the Gigabit to Gigabit path to make it appear super slow. Now the Router will put the Serial to Serial line in the routing table and put the Gigabit to Gigabit line in the Topology table.

theodoxa

mikeybinec wrote: »

Hey Theo--just a quick question that is somewhat related to the OPs question.

Sticking with my diagram I posted a few messages ago--when I run the sh ip ro command I see load balancing to the loopback interface. Now, if I change the bandwidth on the left router, say bandwidth 64, why doesn't the right side peer adjust to the change in bandwidth--Kinda like one side is half duplex and the other is full duplex. If I lower the BW on the left side router, that path/route will be dropped from the sh ip ro table. But the right side router is still at it's configured BW..

Do I make sense?

Thanks

My understanding is that you would have to manually set the bandwidth, because serial doesn't auto detect bandwidth. EIGRP Update packets contain all the metric components as reported by the neighboring router. The receiving router would add the delay for the interface it was received on to the delay component and compare the bandwidth configured on the interface it was received on to the bandwidth in the update packet. It would use the smaller of the two bandwidths (reported vs. configured on its interface).

So, let's say the left router received an EIGRP Update from the right router, the least bandwidth would be 8 Gbps and the cumulative delay would be 5000 usec, which are the figures for the Loopback interface on the right router. The right router needs only send the packet to the Loopback interface to reach that subnet.

256 ((10 Gbps / 8 Gbps) + 500) = 128,256 (Reported Distance)

64 Kbps < 8 Gbps, so when calculating the Feasible Distance the left router would use the smaller number (64 Kbps) and add the delay from the interface the EIGRP update was received on to the delay reported in the EIGRP update.

256 ((10 Gbps / 64 Kbps) + 500 + 2000) = 40,640,000 (Feasible Distance).

[EDIT] The bandwidth and delay are taken from the interface that the EIGRP update is received on, so the bandwidth and delay configured on the right router would only affect routes received from the left router and not routes it sent to the left router. Try adding a Loopback interface to the left router and advertising it into EIGRP. You should find that the right router will show a different Feasible Distance to that Loopback than the left router shows for the Loopback on the right router.

theodoxa

I changed the Bandwidth on Left Serial 0/1/0 from 1 Mbps to 64 Kbps. The Left Router now sees both links as having a least bandwidth of 64 Kbps, so both links have the same FD.

Left#sh ip route eigrp

192.168.200.0/32 is subnetted, 1 subnets
D 192.168.200.1 [90/40640000] via 10.0.0.2, 00:04:16, Serial0/0/0
................[90/40640000] via 10.0.0.6, 00:00:04, Serial0/1/0

I never set the variance on the right router, so it is not performing unequal cost load balancing. Only the route over S0/1/0 (1 Mbps) is used.

Right#sh ip route eigrp

192.168.201.0/32 is subnetted, 1 subnets
D 192.168.201.1 [90/3139840] via 10.0.0.5, 00:04:49, Serial0/1/0

But, if I view the EIGRP Topology Table

P 192.168.201.1/32, 1 successors, FD is 3139840
.........via 10.0.0.5 (3139840/128256), Serial0/1/0
.........via 10.0.0.1 (40640000/128256), Serial0/0/0

I See that it has a lower Feasible Distance for S0/1/0 than S0/0/0. This is because it is using the bandwidth configured on its own S0/1/0, which is still set to 1 Mbps.