From: Pavel Bykov (slidersv@gmail.com)
Date: Wed Mar 18 2009 - 08:46:47 ART
Uh... no, not junk.
There is no perfect load balancing algorithm (that I know of).
MLPPP, Etherchannel, CEF, you name it. Eveything has it's drawbacks.
Some of them are obvious, like differences in flow nature. Some of them are
more subtle, like polarization of ether channels.
Remember, one flow is never a good reference.
On the other hand, if you have 20000 flows, then you'll see that load
balancing doesn't work that bad.
On Wed, Mar 18, 2009 at 12:49 AM, <ron.wilkerson@gmail.com> wrote:
> I understand all that but unless there is some background communication
> between cef and bgp, the load sharing based on bandwidth isn't possible.
>
> Even if cef distributes the buckets to a proper ratio, based on the
> bandwidth, that still doesn't mean that the multiple paths will be used
> according to cef's ratio.
>
> Using your example, if one of the 5 hashes, given to the slower link
> generates more traffic than the faster link, then this whole load sharing
> based on bandwidth doesn't work.
>
> The only way for this feature to work properly is for bgp and cef to
> monitor the interface utilization, and along with that info, decide which
> path to take. I know this doesn't happen, so I think this feature is junk.
>
> ------------------------------
> *From*: Pavel Bykov
> *Date*: Tue, 17 Mar 2009 23:23:34 +0100
>
> *To*: <ron.wilkerson@gmail.com>
> *Subject*: Re: BGP unequal load balancing
> By estimation.
> I'm not sure exactly how it would be done in this case, but CEF balances
> per SRC/DST pair by default.
> To see what CEF choose some IOSes display the actual internal CEF table,
> with 16 CEF entries for every destination network.
> On most IOSes you can see CEF's choice using "show ip cef exact-route
> SRC-ADDR DST-ADDR" command.
>
> If the ration is say 20:5, then out of 16 CEF entries, 4/5 will be sent to
> one link and 1/5 to the other. Every SRC/DST pair generates a hash - from 1
> to 16. In this case 3 hashes are assigned to one interface, and 13 to the
> other.
> So there is a bigger chance that a flow will be assigned to that interface
> with bigger hash probability.
>
>
>
>
>
> On Tue, Mar 17, 2009 at 10:55 PM, <ron.wilkerson@gmail.com> wrote:
>
>> Then I don't see how this feature will work...
>> And bgp is not going to keep track of bandwidth utilization....so how does
>> this work exactly...
>>
>> ------------------------------
>> *From*: Pavel Bykov
>> *Date*: Tue, 17 Mar 2009 22:33:06 +0100
>> *To*: <ron.wilkerson@gmail.com>
>> *Subject*: Re: BGP unequal load balancing
>> No, it's still flow based till you switch it. It will TRY to load balance.
>> By no means it means that it will be successful.
>>
>> On Mon, Mar 16, 2009 at 4:45 PM, <ron.wilkerson@gmail.com> wrote:
>>
>>> Follow up question...
>>> Always wondered how bgp's unequal load sharing worked with cef. Since
>>> the traffic count will be unequal when using this feature, is the router now
>>> performing per packet load sharing?
>>>
>>> Example, with two paths (unequal)
>>>
>>> The traffic count for 1st will be, say 5.
>>> The 2nd is 3.
>>>
>>> Wouldn't this lead one to believe that the router is load sharing per
>>> packet instead of the default flow based?
>>>
>>>
>>> -----Original Message-----
>>> From: "Jared Scrivener" <jscrivener@ipexpert.com>
>>>
>>> Date: Mon, 16 Mar 2009 11:25:28
>>> To: 'Ali El Moussaoui'<mousawi.ali@gmail.com>; 'Cisco certification'<
>>> ccielab@groupstudy.com>
>>> Subject: RE: BGP unequal load balancing
>>>
>>>
>>> Hey Ali,
>>>
>>> Try using the "dmzlink-bw" command on your neighbor statements and the
>>> "bgp
>>> dmzlink-bw" command within the BGP process. Also, you'll need to set
>>> "maximum-paths" to 2 (or higher)from within the BGP process (to enable
>>> load-balancing).
>>>
>>> Cheers,
>>>
>>> Jared Scrivener CCIE3 #16983 (R&S, Security, SP), CISSP
>>> Technical Instructor - IPexpert, Inc.
>>> Telephone: +1.810.326.1444
>>> Fax: +1.810.454.0130
>>> Mailto: jscrivener@ipexpert.com
>>>
>>>
>>> -----Original Message-----
>>> From: nobody@groupstudy.com [mailto:nobody@groupstudy.com] On Behalf Of
>>> Ali
>>> El Moussaoui
>>> Sent: Monday, 16 March 2009 11:06 AM
>>> To: Cisco certification
>>> Subject: BGP unequal load balancing
>>>
>>> Hello Experts,
>>> Is there a method *unequal* Load Sharing When Dual-Homed to One
>>> Internet
>>> Service Provider (ISP) Through a Single Local Router?
>>>
>>> I have 2 links to the Service Provider (20MB and 6MB). I am recieving the
>>> full internet routing table from both links I need to load balance with a
>>> ration of 20 : 6 is that possible?
>>>
>>>
>>> Blogs and organic groups at http://www.ccie.net
>>>
>>> _______________________________________________________________________
>>> Subscription information may be found at:
>>> http://www.groupstudy.com/list/CCIELab.html
>>>
>>>
>>> Blogs and organic groups at http://www.ccie.net
>>>
>>> _______________________________________________________________________
>>> Subscription information may be found at:
>>> http://www.groupstudy.com/list/CCIELab.html
>>>
>>>
>>> Blogs and organic groups at http://www.ccie.net
>>>
>>> _______________________________________________________________________
>>> Subscription information may be found at:
>>> http://www.groupstudy.com/list/CCIELab.html
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>>
>>
>>
>> --
>> Pavel Bykov
>> ----------------
>> Don't forget to help stopping the braindumps, use of which reduces value
>> of your certifications. Sign the petition at
>> http://www.stopbraindumps.com/
>>
>
>
>
> --
> Pavel Bykov
> ----------------
> Don't forget to help stopping the braindumps, use of which reduces value of
> your certifications. Sign the petition at http://www.stopbraindumps.com/
>
-- Pavel Bykov ---------------- Don't forget to help stopping the braindumps, use of which reduces value of your certifications. Sign the petition at http://www.stopbraindumps.com/Blogs and organic groups at http://www.ccie.net
This archive was generated by hypermail 2.1.4 : Mon Apr 06 2009 - 06:44:05 ART