Unequal load balance and OSPF

[biplobkhan]

Hi

Eigrp, ospf, BGP all are support equal load balance.

But if I have HO and 3 branches . Ho connected all branches 3 ISP line.

If 3 ISP is unequal metric and if I run OSPF then data is transfer only one path.

that means no load balancing among the path.I know with the EIGRP can be done Unequal load balance. so with OSPF its not possible.

pls easily explain .

Thanks

Biplob

[tekha]

Not possible with ospf.

As far as I know you can only do unequal loadbalance with EIGRP, and with a combination of e-BGP and static routes.

[bvsnarayana03]

OSPF uses COST of an interface as metric, which is calculated based on BW on interface by default.

But you can directly manipulate the cost of an interface without directly varying BW of interface which hampers QOS & other tools. Please note, lower cost interface is preferred over hisger cost.

So you may want to change the cost on interface over which you need to load-balance.

[biplobkhan]

Hi

My question was if IF ISP line with 64, 128,256 connected branch then when OSPF ruuning among the three path data provide or only one path data pass.

thanks

Biplob

[Kevin Dorrell]

You will only get multiple routes if the total path costs to the destination are equal. Practically, in your situation you will only get multiple paths if you can persuade them they have the same cost. Then the load balancing will be even, not unevan as you are looking for.

Kevin Dorrell

Luxembourg

[biplobkhan]

HI

so In this case eigrp use will be best ?

Biplob

[Kevin Dorrell]

I would say so, as long as all the routers are Cisco.

Kevin Dorrell

Luxembourg

[paolo bevilacqua]

Kevin,

keep in in mind, uneven load balancing with EIGRP requires process switching. In most cases, that is not a viable option.

So generally speaking EIGRP unequal load balancing must be avoided and one has to find other ways to balance traffic.

[Kevin Dorrell]

Sorry Paolo, but I beg to differ. I think the load balancing is handled by CEF. The algorithm is based on a weighted 16-slot hash profile. Here are some documents about it:

http://www.cisco.com/en/US/tech/tk827/tk831/technologies_tech_note09186a0080094806.shtml

http://www.cisco.com/en/US/products/hw/modules/ps2033/prod_technical_reference09186a00800afeb7.html

http://www.cisco.com/en/US/tech/tk365/technologies_tech_note09186a008009437d.shtml

Kevin Dorrell

Luxembourg

[paolo bevilacqua]

Hi Kevin,

this has been discussed many times in the past already.

I've challenged anybody to produce either:

a) One production example of EIGRP unequal load balacing - no matter whta the switching path used.

or

b) A lab example of EIGRP unequal load balance with CEF switching.

So far, nobody has responded. If you can be the one to prove that A and/or B are real, I'll be glad to declare myself wrong.

[Kevin Dorrell]

OK, I'll try it out in the lab and see what happens. It may not be this week though.

Could you give me some links to the previous discussions so that I can read "the story so far".?

Thanks.

Kevin Dorrell

Luxembourg

[paolo bevilacqua]

In this forum, but I've not a ready link.

However the only thing that came out was that nobody at cisco could demosntrate the fact that you can have unequal and CEF (or orvef fast) switching at the same time.

[Harold Ritter]

Paolo,

The eigrp variance feature will work in conjunction with cef switching to load-balance the traffic unequally among outbound interfaces.

routerD#sh ip ro 192.168.13.0

Routing entry for 192.168.13.0/24

Known via "eigrp 1", distance 90, metric 307200, type internal

Redistributing via eigrp 1

Last update from 192.168.23.3 on Ethernet1/0, 00:00:55 ago

Routing Descriptor Blocks:

* 192.168.12.1, from 192.168.12.1, 00:00:55 ago, via Ethernet0/0

Route metric is 307200, traffic share count is 2

Total delay is 2000 microseconds, minimum bandwidth is 10000 Kbit

Reliability 255/255, minimum MTU 1500 bytes

Loading 1/255, Hops 1

192.168.23.3, from 192.168.23.3, 00:00:55 ago, via Ethernet1/0

Route metric is 614400, traffic share count is 1

Total delay is 4000 microseconds, minimum bandwidth is 5000 Kbit

Reliability 255/255, minimum MTU 1500 bytes

Loading 1/255, Hops 1

routerD#sh ip cef 192.168.13.0 internal

192.168.13.0/24, version 25, epoch 0, per-destination sharing

0 packets, 0 bytes

tag information set, all rewrites owned

local tag: 32

via 192.168.12.1, Ethernet0/0, 0 dependencies

traffic share 2

next hop 192.168.12.1, Ethernet0/0

valid adjacency (0x0150DE98)

tag rewrite with Et0/0, 192.168.12.1, tags imposed {}

via 192.168.23.3, Ethernet1/0, 0 dependencies

traffic share 1

next hop 192.168.23.3, Ethernet1/0

valid adjacency (0x0150DBC8)

tag rewrite with Et1/0, 192.168.23.3, tags imposed {}

0 packets, 0 bytes switched through the prefix

tmstats: external 0 packets, 0 bytes

internal 0 packets, 0 bytes

Load distribution: 0 1 0 1 0 1 0 1 0 1 0 0 0 0 0 (refcount 1)

Tags distribution: 0 1 0 1 0 1 0 1 0 1 0 0 0 0 0 (addr 1C135F8)

Hash OK Interface Address Packets Tags imposed

1 Y Ethernet0/0 192.168.12.1 0 none

2 Y Ethernet1/0 192.168.23.3 0 none

3 Y Ethernet0/0 192.168.12.1 0 none

4 Y Ethernet1/0 192.168.23.3 0 none

5 Y Ethernet0/0 192.168.12.1 0 none

6 Y Ethernet1/0 192.168.23.3 0 none

7 Y Ethernet0/0 192.168.12.1 0 none

8 Y Ethernet1/0 192.168.23.3 0 none

9 Y Ethernet0/0 192.168.12.1 0 none

10 Y Ethernet1/0 192.168.23.3 0 none

11 Y Ethernet0/0 192.168.12.1 0 none

12 Y Ethernet0/0 192.168.12.1 0 none

13 Y Ethernet0/0 192.168.12.1 0 none

14 Y Ethernet0/0 192.168.12.1 0 none

15 Y Ethernet0/0 192.168.12.1 0 none

refcount 6

routerD#

Regards,

[paolo bevilacqua]

Harold,

great. What i'd also like to see, is a snapshot with populated statistics:

0 packets, 0 bytes switched through the prefix

tmstats: external 0 packets, 0 bytes

internal 0 packets, 0 bytes

I mean, good so far, now what happens when you pass traffic ?

Any report of customers having this in production ?