[j-nsp] load balancing between multiple BGP links
Robert Raszuk
raszuk at cisco.com
Wed Sep 14 08:53:54 EDT 2005
First Joe has not even indicated if in his OSPF net he runs IBGP or not.
If not then the local pref tips may not turn to be very useful :)
As to the loadbalancing from R4 point of view outbound sure EBGP
multipath has been supported for years - no need for parallel links like
Guy says.
One advice I could add (since cisco-nsp mail alias has been cc-ed :)) is
to use src-dst based hashing (cisco default) rather then just
destination based one.
Using the latter in the case when you have very popular server being
accessed by many src hosts results in all of them traversing the same
link R4-R2 or R4-R1 :(((
The below may also be the caused by poor choice of hashing algoryth on R4:
>>>The three links between our network and Service
>>>provider network are all 1Gbps. Now, we noticed that
>>>load on link R1-R3, R1-R4 is about 50% (in/out), but
>>>load on R2-R4 are about 90% ( in/out).
Cheers,
R.
> Hi,
>
> Guy is not 100% right. Moder BGP implementation allows for hack to ignore last criteria in path selection algorythm.
> This is "BGP Multi-path". This is supported on both JUNOS and IOS.
>
> If you run this command on R4, then trafic leaving your AS by R4 will be distributed over 2 links. Additionaly you has to configure
> alghorytm used to load-balancing - per destination-perfix or per flow.
> Of course traffic leaving your AS by R3 will use only R3-R1 link.
>
> So You can't automaticaly guarantee that all 3 links will be used equaly. You can simply go to situatin when R4-R1 and R4-R2 will be
> loaded equaly, byt R3-R1 can be loaded on different level.
>
> To play with this, You has to paly with BYP attributes per prefix/prefix set:
> Local Preference for play with output traffic.
> MED and AS path for input traffic.
>
> Note.1 Because SP is autonomous system their policy and BGP setup has to be tunned also. There is no way to guarantee
> expected load balancing behavior of SP routers by configuring Customer routers. (e.g. is R3, R4 will use MED,
> then SP can overwrite them by use of Local Preference)
>
> So for output traffic, using BGP multi-path you can reach:
> R3-->R1 50%
> R4-->R1 70%
> R4-->R2 70%
>
> And then play with policy/roadmat on R3 to set-up Local Preference >100 to some set of prefixes (e.g. some AS) and annouce this to R4.
> In effect R4 will send packet addresed to this prefixes to R3 to use R3->R1 link.
>
>
>
> Rafal Jan Szarecki JNCIE #136
> Senior Consultant - Datacom Networks
> Ericsson Poland EPO/S/D
> Office: +48 22 6916635
> ECN: 837 6635
> Mobile: +48 602418971
> Skype: callto://Rafal_Szarecki <callto://Rafal_Szarecki/>
>
>
>
>
>
>
>>-----Original Message-----
>>From: juniper-nsp-bounces at puck.nether.net
>>[mailto:juniper-nsp-bounces at puck.nether.net]On Behalf Of Guy Davies
>>Sent: Wednesday, September 14, 2005 12:50 PM
>>To: Joe Shen; juniper-nsp at puck.nether.net
>>Cc: cisco-nsp at puck.nether.net
>>Subject: RE: [j-nsp] load balancing between multiple BGP links
>>
>>
>>Hi Joe,
>>
>>It can't with the physical layout you described. BGP selects
>>one next-hop as the best (and only one). The only way to
>>load share is to have two (or more) parallel links between
>>just two routers. You then create a static route on each end
>>via each physical next-hop to the loopback on the far end
>>router. Then, you setup an ebgp multihop session between the
>>loopbacks on the two routers. That then uses IGP load
>>sharing to balance traffic across the two parallel links.
>>
>>Other than that, you're stuck with crude methods of dividing
>>up the destination prefixes to approximate load sharing. You
>>could set localpref on routers based on AS path length or
>>some other totally arbitrary parameter to ensure that you're
>>getting something approximating 50/50. But that's pretty
>>horrible, pretty crude and rarely gives stable results.
>>
>>Rgds,
>>
>>Guy
>>
>>
>>>-----Original Message-----
>>>From: juniper-nsp-bounces at puck.nether.net
>>>[mailto:juniper-nsp-bounces at puck.nether.net] On Behalf Of Joe Shen
>>>Sent: 14 September 2005 11:28
>>>To: juniper-nsp at puck.nether.net
>>>Cc: cisco-nsp at puck.nether.net
>>>Subject: [j-nsp] load balancing between multiple BGP links
>>>
>>>Hi,
>>>
>>>How could load on multiple BGP peer links be balanced
>>>automatically?
>>>
>>>The situation we are facing:
>>>
>>> ---------------------------|
>>> | Service provider |
>>> | |
>>> --R1--------------------R2---
>>> |\ |
>>> | \ E-BGP |
>>> | \ ----------------\ |
>>> | \|
>>> ----R3--------------------R4-|
>>> | Our Network |
>>> | (OSPF) |
>>> -----------------------------
>>>
>>>The three links between our network and Service
>>>provider network are all 1Gbps. Now, we noticed that
>>>load on link R1-R3, R1-R4 is about 50% (in/out), but
>>>load on R2-R4 are about 90% ( in/out).
>>>
>>>How could we balance those load on the three links
>>>automatically? or must we tune the route mannually?
>>>
>>>Is there any technical guide on this?
>>>
>>>thanks in advance.
>>>
>>>Joe
>>>
>>>
>>>
>>>
>>>
>>>
>>>___________________________________________________________
>>>雅虎免费G邮箱－中国第一绝无垃圾邮件骚扰超大邮箱
>>>http://cn.mail.yahoo.com
>>>
>>>_______________________________________________
>>>juniper-nsp mailing list juniper-nsp at puck.nether.net
>>>http://puck.nether.net/mailman/listinfo/juniper-nsp
>>>
>>
>>This e-mail is private and may be confidential and is for the
>>intended recipient only. If misdirected, please notify us by
>>telephone and confirm that it has been deleted from your
>>system and any copies destroyed. If you are not the intended
>>recipient you are strictly prohibited from using, printing,
>>copying, distributing or disseminating this e-mail or any
>>information contained in it. We use reasonable endeavours to
>>virus scan all e-mails leaving the Company but no warranty is
>>given that this e-mail and any attachments are virus free.
>>You should undertake your own virus checking. The right to
>>monitor e-mail communications through our network is reserved by us.
>>
>>
>>
>>_______________________________________________
>>juniper-nsp mailing list juniper-nsp at puck.nether.net
>>http://puck.nether.net/mailman/listinfo/juniper-nsp
>>
>
>
> _______________________________________________
> juniper-nsp mailing list juniper-nsp at puck.nether.net
> http://puck.nether.net/mailman/listinfo/juniper-nsp
More information about the juniper-nsp
mailing list