iBGP Multipath Load sharing

[net guru]

Hello experts,

this is a basic bgp question ( am not including bgp use in MPLS VPN ) 

why can we use multipath only if the paths have the same attributes : "weight, local preference, AS-path (the entire attribute and not just length), origin code, MED , and IGP  distance." I know it is not accurate to compare bgp with IGPs , but if we consider EIGRP unequal cost load balance compared to BGP that insists on the attributes to be the same , I know that there's a huge aspect that I am missing behind it .

why could it be ?

[MHM Cisco World]

simple answer is 
EIGRP have special method to detect LOOP 
""Note: If a path is not a feasible successor, the path is not used in load balancing. Refer to the Feasible Distance, Reported Distance, and Feasible Successor section of Enhanced Interior Gateway Routing Protocol for more information.""

where the BGP if you override all attribute then there is chance of LOOP.

[paul.driver]

Hello
You can implement unequal load-sharing within bgp also, bgp dmzlink-bw and maximum-paths ibgp x

[Giuseppe Larosa]

Hello @net guru ,

BGP original design does not support multiple paths to a destination and the support of iBGP multipath is a later addition.

The basic idea behind it  is to take advantage of multiple paths if they can be compared , that is they need to have same settings for the BGP attributes local preference AS path, MED origin until the same IGP metric to BGP next-hop.

This is clearly done with the intention to perform Equal cost multi path when possible and when it makes sense.

The requirement about the AS path attribute is specific to BGP as it is stated that not only must be of the same length but the same AS sequence.

This is probably a consistency check so that the peer AS is the same.

As already noted by @MHM Cisco World even EIGRP un equal cost multi path have requirements only feasible successors passing the FC can be used.

Hope to help

Giuseppe

[Joseph W. Doherty]

"I know that there's a huge aspect that I am missing behind it .

why could it be ?"

Possibly, the huge aspect would be how do you split the traffic load, proportionally, between the unequal multiple paths?  (Laugh, ever study EIGRP's full composite metric calculation formula?  Consider doing something with all of BGP's attributes, and, for-the-icing-on-the-cake, consider trying to obtain an industry wide agreement for a standard [one "advantage" with EIGRP, or OER/PfR, they are proprietary].)

Then, once you've a standard for how all the attributes, combine, into some kind of metric weight used for traffic load distribution, you have the issue the additional processing load keeping track of how actual flow loads are meeting the assigned weights and recalculations as BGP attributes dynamically change.  (Ever notice how rarely EIGRP's unequal cost multipath seems to be used?)

[net guru]

So in essence , if I am to translate your reply to a functional response , it is not about , theoretically speaking , feasible to load share traffic across routes with unequal attributes , but in practice it will affect how the flow should be forwarded since we're dealing with relatively complex attributes and calculations that will lead to the famous argument finisher word ( laugh, "Routing LOOP" ) . 
Tl,DR : Routing loop + unexpected forwarding behavior ? 

[Joseph W. Doherty]

If I understood your response to my post, actually it would be theoretically possible, but for practical reasons, not done (for reasons, I suspect it hasn't been done for IGP excepts for EIGRP).

Further, with more modern technologies, like OER/PfR, which can provide multi-path load balancing, including for unequal paths, including over BGP, it sort of makes the question of doing such (unequal) multi-path with BGP moot, at least in my opinion.

In fact having used OER, on the Internet, when Cisco first released it, it sort of made having complete Internet route tables on my routers moot too.  I found using OER with just default routes to the Internet worked much "better" than having complete Internet route tables.

[MHM Cisco World]

https://www.cisco.com/c/en/us/support/docs/ip/border-gateway-protocol-bgp/13753-25.html
this link explain the BGP best path, 
you can see for multipath all first 6 attribute must match 
and then after point 6 you can config multipath. 
i.e. this already cover by cisco and mention that not all attribute must equal but the most important attribute must equal to config multipath. 

[MHM Cisco World]

friend only share
show ip bgp x.x.x.x
and I will inform you the command you can use to override the BGP best path selection