[c-nsp] T1 Bonding with PA-MC-T3

[Ben Steele]

Actually I can vouch for per-packet working fine for sensitive  
applications like VoIP as long as your bonded lines are basically  
parallel in the sense they are the same technology over the same  
distance with the same characteristics, ie multiple T1 lines through  
the same carrier to the same destination, most VoIP software these  
days has allowance built in for de-jittering and buffering of packets  
so an out of order packet shouldn't actually matter that much over  
this particular scenario, it's more when the lines you are per-packet  
sharing over are quite unbalanced in their characteristics ie latency/ 
serialization/throughput etc.

In my real world experience per-packet is fine for bonded T1 based on  
the above guidelines.

Ben
On 16/03/2008, at 1:39 AM, Nick Voth wrote:

>>>> Thanks very much David. That definitely helps. Yes, our 2 T's are  
>>>> on the
>>>> same path to the destination so it looks like per-packet would be  
>>>> best in
>>>> that case. However, with "per-packet" can you utilize the full  
>>>> speed of the
>>>> 2 T's as if they were bonded like in MLPPP? That's the ultimate  
>>>> goal here.
>>>
>>> Memory says "yes" you get the full speed of 3Mb available with per- 
>>> packet.
>>> It has been 8 years since I last used it, but I'm sure someone can  
>>> back
>>> up my memory. ;)
>>
>> per-packet will give you a 3mb pipe...but the advantage to MLPPP is  
>> it
>> does the same and guarantees preservation of packet ordering.  Per- 
>> packet
>> can/will result in out of order packets which can really foul up  
>> VOIP.
>>
>> ----------------------------------------------------------------------
>>  Jon Lewis                   |  I route
>>  Senior Network Engineer     |  therefore you are
>>  Atlantic Net                |
>> _________ http://www.lewis.org/~jlewis/pgp for PGP public  
>> key_________
>
> Thanks again guys. It seems that the consensus from all that I've  
> read and
> the replies that I've gotten is that using CEF with "per-packet" is  
> a great,
> low overhead way to do it, BUT with sensitive applications like  
> VoIP, MLPPP
> will keep your packets in order better. I'll experiment and see.  
> Thanks
> again for everyone's advice!
>
> -Nick Voth
>
>
> _______________________________________________
> cisco-nsp mailing list  cisco-nsp at puck.nether.net
> https://puck.nether.net/mailman/listinfo/cisco-nsp
> archive at http://puck.nether.net/pipermail/cisco-nsp/