BGP DMZ Link bandwidth

[alex_ciobanu]

Hello,

As Cisco official documentation says, The Border Gateway Protocol (BGP) Link Bandwidth  feature is used to advertise the bandwidth of an autonomous system exit  link as an extended community.

If we configure the AS in such a way that even the iBGP peers will know about the 2 possible routes for one prefix, doing unequal cost load balancing, how do they know what is the load of the interface ? They receive at some hello time interval information about the interface utilization through the usage of extended communities ?

Documentation says BGP is doing load balancing, and load balancing means balancing the packets according to interface utilization/load.

Thanks in advance for replying !

[handoko wiyanto]

hi Alex,

who knows you might find this link useful:

http://bit.ly/ayyrKY

http://bit.ly/LebRQn

regards,

[alex_ciobanu]

Thanks, but replying to a question with 'read the RFC' or an INE technical tutorial is not a good answer you know...

[handoko wiyanto]

are you having trouble with your bgp config, or you just learn the feature for certification?

[smehrnia]

Hi Alex,

If we configure the AS in such a way that even the iBGP peers will know about the 2 possible routes for one prefix, doing unequal cost load balancing, how do they know what is the load of the interface ? They receive at some hello time interval information about the interface utilization through the usage of extended communities ? 
Documentation says BGP is doing load balancing, and load balancing means balancing the packets according to interface utilization/load.

First, Load Balancing does NOT mean balancing the packets according to interface Load, as you said above.

it means Balancing the Traffic Load across multiple paths.

Now your answer:

The BGP Link Bandwidth feature allows BGP to be configured to send traffic over multiple iBGP or eBGP learned paths where the traffic that is sent is proportional to the bandwidth of the links that are used to exit the autonomous system. well, in this scenario there's nowhere that a link's load has effect on routing.

plz Rate if it helped.

Soroush,

[alex_ciobanu]

Load balancing means balancing the packets according to interface load/utilization. Balancing the traffic load accross different multiple paths is called load sharing. Load balancing and load sharing are 2 things people usually don't differentiate and come to wrong definition. There is one video by this guy who is one of the few 5xCCIE and he is explaining the difference between load sharing and load balancing, even though the video is not related to BGP -->  http://goo.gl/L84ag. Same confusion also takes place for EIGRP, where the definition calls it unequal cost load balancing, but when you check the routes you will see there one line saying "traffic share count".

Now according to Cisco documentation,  "This feature is used with BGP multipath features to configure load balancing over links with unequal bandwidth. " Since load balancing is balancing the traffic according to interface load, back to the beginning, my question still remains: how do the peers know about the interface load of other peers(especially iBGP peers).

[handoko wiyanto]

hi Alex,

the bandwidth terms on internal routing protocol or BGP are only used for the metric. with this metric, the routing protocol can take the decision (routing), it doesnot take the actual troughput information from the interface.

i believe the terms load balancing on cisco docs according to this feature is reffering to load sharing. bgp dont do load balancing.

if what youre looking for is more "concept" question when you learning something, i would recomend http://learningnetwork.cisco.com, there you can find good answers. if you having a real issue with bgp on your production router, i believe this forum is the best place for looking answers.

regards,

[alex_ciobanu]

Hi handoko,

This is what I also think, that there is just a gap in the documentation, and they should reffer to load sharing the traffic, not load balancing. They should call it load sharing based on configured bandwidth of the links to peers.

[smehrnia]

btw, Load balancing means, balancing the traffic sharing according to a measure i.e.: Bandwidth. not necessarily LOAD of THE LINK.

   Benefits of the BGP Link Bandwidth Feature

The BGP Link Bandwidth feature allows BGP to be configured to send traffic over multiple iBGP or eBGP learned paths where the traffic that is sent is proportional to the bandwidth of the links that are used to exit the autonomous system. The configuration of this feature can be used with eBGP and iBGP multipath features to enable unequal cost load balancing over multiple links. Unequal cost load balancing over links with unequal bandwidth was not possible in BGP before the BGP Link Bandwidth feature was introduced.

http://www.cisco.com/en/US/docs/ios/12_2t/12_2t2/feature/guide/ftbgplb.html#wp1058357

plz Rate if it helped,

Soroush.

[alex_ciobanu]

Ok guys.. I am trying to get a real and serious answer in here, we all know you're trying to get as many points as possible, by looking at your other posts, but answering to questions with copy-paste and official documentation links is not that helpful.

I am still waiting for other opinions from some advanced users.

[Giuseppe Larosa]

Hello Alexandru,

the dynamic load on the interfaces is not taken in account by this BGP feature and by any real world IP routing protocol.

The simple reason is that it would introduce network instability: oscillations with never ending recalculations and changes in metric. If the metric would change with load it would be a nightmare.

Even EIGRP implementation was not able to deal with dynamic change of the load on interfaces and if the K value for load would be enabled the load value would be that of the interface at the moment EIGRP started to run on it, and it  would not change over time, not reflecting real load on the interface.

This aspect of EIGRP implementation had been explained here in the forums some years ago by Cisco expert Russ White.

see

https://supportforums.cisco.com/message/792452#792452

Even if now code may allow for dynamic load support, real world EIGRP implementations use only the delay and bandwidth components of metric ( default K values).

BGP is an EGP and its level of abstraction is high. BGP metric MED is only one of many factors that are used to choice the BGP best path. I usually say that IGP have a scalar mono-dimensional metric and BGP uses a multi dimension array of "metric values" with a clear hierarchy of criteria. (BGP path vector)

So we cannot expect BGP MED or a BGP community to reflect dynamic load on interfaces.

From a technical point of view is not feasible.

You are right that there is confusion between load sharing and load balancing and most of Cisco documentation use them as they were the same.

Attention to details and terminology is important, but it is also important to develop a pragmatic approach.

Hope to help

Giuseppe

[alex_ciobanu]

Hello Giuseppe,

Yes I agree with you, load balancing and dynamic protocols is not a scalable approach, even if the word 'dynamic' comes into place.

Regarding EIGRP, they just call it wrong 'load balancing', and the proof stays in the output of 'show ip route' that states the "traffic share count xxx" which tells directly that the router is doing load sharing, and not load balancing.

Getting to BGP I would say I see the same thing, they say its load balancing proportional to the bandwidth of the exit links, but all it does it just injects routes with the metric calculated proportional to those bandwidths, and then the router will just load share the traffic accordingly.