Variance

[dathaide]

hi

What will my variance be in this case. I set it to 2 but don't eigrp load balancing across unequal paths

here is the scenario. I have two routers at the lab. Each has two 64K serial links. One serial link is directly connected to the second router. The second serial link is connected via a satellite provider.

Routing Descriptor Blocks:

xx.xx.xx.xx (Serial0/0.1), from x.x.x.x, Send flag is 0x0

Composite metric is (40515072/28672), Route is Internal

Vector metric:

Minimum bandwidth is 64 Kbit

Total delay is 20120 microseconds

Reliability is 255/255

Load is 1/255

Minimum MTU is 1500

Hop count is 3

x.x.x.x (Serial0/1.1), from x.x.x.x, Send flag is 0x0

Composite metric is (41027072/40515072), Route is Internal

Vector metric:

Minimum bandwidth is 64 Kbit

Total delay is 40120 microseconds

Reliability is 255/255

Load is 15/255

Minimum MTU is 1500

Hop Count 4

[samsam_wang]

conf t

int se 0/0.x

delay 2000

(please add 'delay 2000' two sides)

now the metric 40515072 should change to 41027072

that is it, if it doesn't work, give me the new metric value after you add 'delay 2000'.

[dathaide]

thank u

for my benefit could u please post the equation with the variables. This delay must be added at both the end. thanks

[ruwhite]

http://www.cisco.com/warp/customer/103/eigrp-toc.html#eigrpmetrics

Is what you're after... The delay isn't added on both ends, but the total metric does have to be less than the feasible distance for the route to be considered loop free, and thus to be used as an unequal cost load share in the routing table. You also have to be careful of the math, do it in the right order, and round in the right places.

:-)

Russ.W

[dathaide]

hi

I used the formula

metric - [(10^7/64)*256]= total delay *256

Thus using this formula i get 40512000 slight shy of my delay of 40515072

thanks

however i get Total delay of 4012 when i use

metric= bandwidth + delay and solve for the delay

41027072 = (1000000/64)*256 + 256*total delay

total delay is 4012

Please verify

Thus my show ip eigrp top should read the following. Since i have not tested this in a lab i assume that my reported distance is less than my feasible distance

Thus a

int s0/0.1

delay 4012

Should work and eigrp should load balance across the links

xx.xx.xx.xx (Serial0/0.1), from x.x.x.x, Send flag is 0x0

Composite metric is (41027072/28672), Route is Internal

Vector metric:

Minimum bandwidth is 64 Kbit

Total delay is 20120 microseconds

Reliability is 255/255

Load is 1/255

Minimum MTU is 1500

Hop count is 3

x.x.x.x (Serial0/1.1), from x.x.x.x, Send flag is 0x0

Composite metric is (41027072/40515072), Route is Internal

Vector metric:

Minimum bandwidth is 64 Kbit

Total delay is 40120 microseconds

Reliability is 255/255

Load is 15/255

Minimum MTU is 1500

Hop Count 4

[ruwhite]

It's actually:

10000000/64 = 156250

156250 + 4012 (total delay) = 160262

(if you stopped here, this would be the IGRP metric--the EIGRP metric is the IGRP metric pushed left by 8 bits, so multiply by 256)

160262 * 256 = 41027072

This should be right. Based on what you are showing in the output above, though, the two routes are equal cost, not unequal cost (?). Do you have actual output from the router? The left number is the distance through this path, the right is the reported distance, which is our neighbor's cost to reach the destination. The right number of the alternate paths has to be less than the left number of the best path.

Anyway, I'd verify the metrics on a router. If this output is right, you shouldn't need variance here, I don't think.

:-)

Russ.W

[samsam_wang]

i think variance doesn't work on ethernet or fastethernet, so i only use delay

and (metric1-metric2)/256 is the delay value.

99% it will be right, if not, i can get the right value at the second time.

[ruwhite]

Variance shouldn't have anything to do with the media type (?).... If there are two routes, both provable loop free (both feasible successors, or rather the successor and a feasible successor), and the total cost through one is less than the best path times the variance, both routes are installed in the routing table. There isn't a check for the media type.

:-)

Russ.W

[samsam_wang]

hi,

you got me wrong.

the range of variance is from 1 to 128, just see the topology.

R1----64k-----R2

R1---1544k----R2

R1----10M-----R2

run eigrp

the variance doesn't work for the topology.

only delay works

I add the below delays, and 3 metircs are equal(=40537600)

add delay of 149774 for 1544 link

add delay of 157250 for 10M link

tell me why the variance range is 1-128

Thanks

[ruwhite]

Could you post the actual metrics of a route learned through these three links in this case?

:-)

Russ.W

[samsam_wang]

R1 ----64K-----R3

R1 e0----10M-----e0 R3

R1---1544K----R2

R2---1544K----R3

now, from R1 to E1 of R3, there are 3 routes, the metrics are,

first,

(10000000/10000+(1000+1000)/10)*256=307200

second

(10000000/64+(20000+1000)/10)*256=40637600

third

(10000000/1544+(20000+20000+1000)/10)*256=2707456

right now, for the metric 307200 and 40637600, the variance doesn't work, because the range is from 1 to 128.

and I add delay 149774 on the link of 1544K,

now the metric is

(10000000/1544+(20000+20000+1000)/10+(149774-2000))*256=40537600

please note, 2000 of (149774-2000)

same as above

I add delay 157250 on the link of 10M

now the three metric are equal.

sometimes i can not explain it clearly because the language. just give me your email, ruwhite, i send the picture to you.

[dathaide]

Just my 2 cents

For variance to work you must have a Feasible successor. Your backup routes may not be Feasible successors and hence variance will not work. To be a feasible successor reported distance of the backup should be less than Feasible distance of the primary route.

thanks

[ruwhite]

Yes, exactly.... Which is why I asked for the actual topo table entries, to see if they were feasible succesors.... :-)

Russ.W