The problem in most instances outside of a co-lo/IX is the LEC. Around
here, AT&T won't deliver a 10G hand-off until you're committed to 4Gb/s.
And a 10Gb port is like $7k minimum.
On 12/14/2016 2:09 PM, Sean Heskett wrote:
In my opinion anytime you need more than 1gbps just order a 10gig
circuit and purchase 1 or 2 gig burstable.
The only time you wouldn't do this is if you want path diversity. But
it sounds like you are looking for capacity not diversity.
Bonding 1gig circuits is not like bonding T-1s. Also with seperate
1gig circuits it's hard to load balance because it's harder to
instruct the traffic coming to your network to use circuit or the
other. You can control how traffic leaves your network but you can't
really control how it gets to you. (There are ways to trick bgp yes
but why bother, just get a burstable 10gig and move on)
2 cents
-Sean
On Wed, Dec 14, 2016 at 10:02 AM Matt <[email protected]
<mailto:[email protected]>> wrote:
> This this one circuit from multiple upstreams? If so, BGP, is
this two 1gig
> circuits from the same upstream? If so, BGP. Let me think,
when in doubt
> BGP. J
So if we have two BGP sessions at this location with them over two
GigE pipes the traffic should balance across both of them? They say
its more cost effective for multiple GigE circuits until you reach ~4
rather then a fractional 10G.
I remember years back load balancing T1 circuits. VOIP hated out of
order packets. Switching to MLPPP made life better there.
>> If you receive two GigE Internet circuits from your upstream how do
>> you load balance them to work as a 2 Gigabit Internet circuit?
They
>> use Juniper and we use Mikrotik.
>>
>> Just wandering what we do after we out grow our single GigE at this
>> location.
>>
>
>
> Load balancing is barf. Try to get a 10G port first. If I was
stuck with
> Nx1G I'd aim for BGP multipath load sharing.
