[c-nsp] Dual OC3 - Separate Carrier - Load-balancing 7201

[Justin M. Streiner]

On Tue, 5 Apr 2011, arulgobinath emmanuel wrote:

> " out of order could be a major headache, and throughput could suffer
> greatly." I haven't tried nor experience in similar setup according to my
> search MLPPP designed to handle the out of order packet (rfc1990)   but it
> can impact the router performance & buffer space.

If you're doing per-packet load-sharing, the only way to handle 
out-of-order packets is to have deep buffers.  Either the routers or hosts 
have to buffer packets to get them back into the right order, or the 
applications themselves need to be able to tolerate packets arriving out 
of order.  In the worst case, the proper sequence of packets does not 
arrive before the buffer fills, which would force the application to 
either re-transmit the affected chunk of data, or throw an error.

None of these events are particularly good for maximizing throughput.

jms

> On Tue, Apr 5, 2011 at 3:38 AM, Justin M. Streiner
> <streiner at cluebyfour.org>wrote:
>
>> On Mon, 4 Apr 2011, Mark Mason wrote:
>>
>>  We are planning on terminating dual OC3 point-to-point circuits
>>> (PA-POS-2OC3) from different carriers on 7201 NPE-G2's at two of our DC's
>>> and do either HDLC CEF per packet load-balancing or multilink PPP bundle
>>> together. What are some of the questions and responses from the field for
>>> anyone who has done this type of setup?
>>>
>>
>> I would stay away from per-packet load-sharing in this design, unless there
>> is a really compelling reason to use it.  The biggest reason to stay away is
>> that OC3 (technically, OC3c) circuits from different carriers to the same
>> locations could have vastly different end-to-end latencies.  If they are
>> significantly different, dealing with packets arriving out of order could be
>> a major headache, and throughput could suffer greatly.
>>
>> You might also want to look at a Gigabit Ethernet solution, if that's an
>> option and you're not chained to a POS design.  The operating costs of a
>> pair of gig-e circuits could very well be substantially lower than a pair of
>> OC3cs, with the added benefit of being able to provide a lot more bandwidth.
>>  If you were to get two gig-e circuits, the caveat above related to
>> per-packet load-sharing would still apply.
>>
>> In a POS world, if you need more than 155 Mb/s, you either need to install
>> another POS circuit, or start upgrading to OC12c or higher.  That also
>> throws in the need to purchase new router hardware, because the 7201 won't
>> handle a POS OC12c, so your costs per megabit wouldn't scale too well.
>>
>> jms
>>
>> _______________________________________________
>> 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/
>>
>