[c-nsp] Multilink Vs Serial Conf.

[Oliver Boehmer (oboehmer)]

Gangasagar Amula <mailto:sagar_cisco at rediffmail.com> wrote on Friday,
July 08, 2005 11:17 AM:

> Hi Oli,
> 
> Pls clarify one more query of mine....
> When we use multilink..is it we have to configure no fair-queue on
> the serial link(s) and fair-queue on the multilink.... 

as said: you need to enable it on the Multilink interface (but I
recommend you use the new MQC to configure any QoS). If you enable it on
the serial link(s), it is ignored and is changed to Fifo:

VXR-oli#sh int mul1 | i Queueing strategy
  Queueing strategy: weighted fair
VXR-oli#sh int s4/0:0 | i Queueing strategy
  Queueing strategy: weighted fair  [suspended, using FIFO]
VXR-oli#sh int s4/0:1 | i Queueing strategy
  Queueing strategy: weighted fair  [suspended, using FIFO]
VXR-oli#

> Again if we are using Multilink for bundling purpose (right now we
> are not using it for bundling purpose as mentioned earlier...) is the
> load sharring between the links happens per packet or per
> destination?   

Well, neither. Load-sharing over Multilink interfaces works
transparantly to the upper layer protocols. PPP multilink uses its own
algorithm to achieve a pretty symmetrical link utilization, and it
guarantees delivery in order by using sequence#..

	oli


> 
> 
> 
> On Fri, 08 Jul 2005 Oliver Boehmer(oboehmer) wrote :
>> 
>>> 
>>> In our organisation.....we are using Multilink only for "fragment
>>> delay" feature of ppp multilink....
>>> 
>>> Hence we have configured one multilink for one link....
>>> 
>>> And we are facing a prb....i.e. if I am giving Fair-queue in
>>> Multilink and serial port...we get Queueing strategy: Weighted
>>> fair... but in serial its "Queueing strategy: weighted fair 
>>> [suspended, using FIFO]" 
>>> 
>>> Is it Multilink configuration prefered first and then Serial Link
>>> configuration....i.e. the fair-queue is really happening and then
>>> the data is thrown to serial as fifo.... Is it fair-queue doesnt
>>> happen twice....for same data.... 
>>> 
>>> Is my analyses to the observation right...?
>> 
>> yes, this is correct. All the queuing happens at the multilink
>> interface, the physical is then just a fifo pipe.. This is the only
>> way to perform fancy queuing on multilink links as one serial has no
>> means to know how the other interface's buffers/queues look like...
>> 
>>      oli
> 
> 
> 
> 
>  <http://clients.rediff.com/signature/track_sig.asp>