From: Richard Gallagher (rgallagh@cisco.com)
Date: Tue Nov 08 2005 - 22:11:34 GMT-3
Carlos G Mendioroz wrote:
> Hmmm,
> I may be way off (more common than would like to) but:
> -I don't know why you say CEF is more CPU intensive.
> AFAIK, it's not supposed to be so.
This should be much less CPU intensive.
> And CEF has a static (given that routing does not change) FIB table
> based on RIB, so per packet load balancing is not an option.
Per-packet is possible with CEF, although it's done on a per flow basis,
so packets in the same flow will always be sent on the same path. So the
more flows you have the more even the sharing of the links will be.
You need to enable:
ltd-6-7507-15(config-if)#ip load-sharing ?
per-destination Deterministic distribution
per-packet Random distribution
ltd-6-7507-15(config-if)#
There are a number of load-sharing algorithms too:
ltd-6-7507-15(config)#ip cef load-sharing algorithm ?
original Original algorithm
tunnel Algorithm for use in tunnel only environments
universal Algorithm for use in most environments
ltd-6-7507-15(config)#
> There is a "full" option where layer 4 data is taken into account for
> the hash that determines the neighbour when many active routes exist in
> the FIB, but that's it.
The L4 option is only available on cat6k AFAIK for the moment.
> Again, AFAIK per packet load balancing requires process switching.
See the following for more info:
http://www.cisco.com/warp/public/105/loadbal_cef.html
Rich
>
> -interface route-caching has some non obvious (to me) semantics.
> It does cache ACL policy (i.e. access-group pass/deny) on the flows,
> thus impacting at least per packet CPU usage. I don't know what other
> consequences it has. And it can be used with CEF.
>
> Hope this helps.
> -Carlos
>
> Holmes @ 08/11/2005 12:55 dixit:
>> Hi all -
>>
>> I'm looking at a load-balancing case, where IP per-packet
>> load balancing is being done. (I'm using cef, and ip load-sharing
>> per-packet).
>>
>> Briefly, if we were doing load balancing over 2 T1 links, router
>> to router, say a router has 2 serial interfaces, if route-cache were
>> enabled, it doesnt always have to do a look up, but would
>> have an entry that for any traffic with this source / destination
>> pair and send all further traffic over this interface (basically
>> per-session or per destination load balancing).
>>
>> With cef, there no such source/destination pair stored, its a
>> little more CPU intensive, but we can get IP packet level
>> load-balancing.
>>
>> Now to my question :
>>
>> Why is `ip route-cache / no ip route-cache configured at the interface
>> level, and not at the global level ? Isnt the decision to send a
>> packet out to a certain destination over a certain interface made
>> before-hand ie. If you configure no ip route-cache at the interface
>> level,
>> isnt it too late, the path / interface has already been chosen.
>>
>>
>> Thanks for any replies.
This archive was generated by hypermail 2.1.4 : Thu Dec 01 2005 - 09:12:05 GMT-3