OSFP Load Share

[hs08]

Hello,

We have 2 router connected to different 2 different ISP with different bandwidth also. This 2 routers also connected to the core switch. How on the core switch, the OSPF will give more traffic to the 1st router (assumed 1st router connected to ISP with high bandwidth), and give more less traffic to the 2nd router?

[Richard Burts]

What you ask is not easy to achieve in OSPF. EIGRP does support unequal load sharing using variance. But the behavior of OSPF is to share equally. There are some alternatives to consider which could achieve unequal traffic flows to the ISPs, such as static routes for some destinations or Policy Based Routing. But with these alternatives there are some challenges, such as what happens if the ISP with more bandwidth is having problems or is down.

[hs08]

can we manipulate the ospf cost?

[MHM Cisco World]

the OSPF only support equal cost path, 
eigrp support unequal path. 

if you modify the cost to be same then you will get 50/50 %.
sorry this OSPF limitation. 

[hs08]

Hi.. so if both link have different cost the active link is only one and other link will be act as backup, right?

[MHM Cisco World]

Yes sorry, one link only, other will be backup. 
this only for OSPF 
you can run EIGRP 
or you can use static route with same AD 
and CEF will do load sharing depend on BW for each link. <<- this need some lab to be sure that share traffic count is different between two path 

[hs08]

Hello,

Interesing talk about ospf bandwidth. I have 2 scenario like this :

- if both ospf cost configured for 100mbps but the actuall bw is 40mbps and 100mbps are this mean link with 100mbps only can reach max 40mbps?

- if both ospf cost configured for 40mbps but the actuall bw is 40mbps and 100mbps are this mean link with 100mbps only can reach max 40mbps?

[MHM Cisco World · Accepted solution]

Hi friend 
I make Cost for both path same Cost 10 but I change the BW make one 1000 and other 2000 
and still same result the share count is 1/1 meaning traffic will share 50/50 %. 

so the Cost is override the BW in load sharing of CEF.

[Joseph W. Doherty]

"I make Cost for both path same Cost 10 but I change the BW make one 1000 and other 2000 
and still same result the share count is 1/1 meaning traffic will share 50/50 %. 

so the Cost is override the BW in load sharing of CEF."

@MHM Cisco World is 100% correct in his observations.  Perhaps, it's worth knowing why.

First, understand, RFC OSPF has a cost metric which has nothing to do with bandwidth.  (One reason why my prior posting kept mentioned adjusting OSPF costs.)

Cisco's implementation of OSPF added auto costing based on "bandwidth", with their default reference bandwidth being 100 Mbps (i.e. 100 Mbps, or "faster" generates the minimum OSPF cost of "1".)

Other network vendors' OSPF implementations might not support any auto costing but if they do, often they use a higher reference bandwidth.

So, it sort of make much sense that explicitly setting OSPF cost "overrides" Cisco's OSPF cost auto calculated from bandwidth.

Since Cisco does support RFC explicit OSPF costing, and auto OSPF costing, which should you use?  Well, as so often the case, the answer is "it depends".  Both approaches have their advantages and disadvantages (which I'll not go into here), and they are not mutually exclusive.

[Joseph W. Doherty]

Interface bandwidth statements, alone, don't have any direct bandwidth control.  Something like a policer or shaper basing its bandwidth limitations on an interface bandwidth statement, can control (limit) bandwidth.  Interface bandwidth, vis-à-vis OSPF, also does not control bandwidth, but in the Cisco OSPF implementation, it can influence which path it chosen.   

[paul driver]

Hello
Just like to add you could also look into Performance Edge routing (PfR) which can calculate best path routing best on link utilization, reachability/latency etc.. which with normal routing best selection isn’t applicable - here

[ToughGuy]

If you have 2 ISP's connected to the core, you can use BGP ( with IPSLA tracking )

not OSPF, you can use EIGRP (UECLB). 

or have a static route with higher AD on the core to manipulate the traffic (if one goes down , the other comes up)

[Joseph W. Doherty]

As many of the other posters have noted, OSPF only supports ECMP, equal cost multiple paths.  I.e. if OSPF sees any cost difference, at all, it will not use higher cost paths, concurrently with lower cost paths.

Can we "lie" to OSPF, i.e. make unequal path have the same cost?  Yep, sure can.  Further, it can work well, even very well if paths are close to being equal.

Other posters have mentioned there are alternatives for actually performing unequal cost, proportional, routing.  They are 100% correct.  However, when discussing routing, load balancing/sharing (including proportional) is mostly done per flow.  What this means in practice, if, for example, you have two links, one running at 100%, and the other at 0%, flow routing may direct a new flow to use the 100% busy link.  Further, if two flows are on the 100% link, while another is idle, i.e. 0%, neither flow is dynamically shifted to the unused link.  So, in practice, often equal cost vs. variable cost routing, or "lying" to the routing protocol that paths are equal, doesn't make a huge difference, unless the paths are very unequal, and in that case, in might make sense to not even use a much "worst" path except in a failure situation.

Years ago, we had a branch that was intended to have obtain two DS1s.  The one DS1, though, was was accidentally ordered as a DS3.  Quite a disparity of bandwidths, 1.5 Mbps vs. 45 Mbps.  Even if we used proportional based routing, do we want any one flow limited to 1.5 Mbps, vs. "sharing" the DS3?  We chose the latter.

We also had many dual DS1s, with one being ATM.  ATM has overhead, such that's its effective bandwidth is a tad less than other technologies on the other DS1.  I.e. by default, using OSPF, it wouldn't use the ATM link unless we "lied" to OSPF, making its cost the same as the other DS1.

Notwithstanding the foregoing, @paul driver has mentioned one of my favorited Cisco technologies, for getting the most out of concurrent links, PfR.  Some if its features, are "advanced", but PfR can, pretty easily, dynamically, and constantly, load balance/share multiple links.  Further it can do this proportionally.  I.e. in the real world case of having a DS1 and DS3, PfR can be configured to keep the load usage the "same" % usage, on multiple paths (based on actual bandwidth capacity of each link).

If you cannot use PfR, you can make your two ISPs routing costs alike, so that OSPF will use them, or keep them distinct so that OSPF will only use the best available, or use another methodology that supports proportional routing distribution.  Again, any proportional routing technique, likely will not make a difference with short term measurements, unless you have lots of concurrent flows and they need/use about the same bandwidth each, but long term measurement may show effect of the over all proportion sharing.