[clug] packet management on multi-homed network

[Daniel Pittman]

On Wed, Dec 29, 2010 at 21:43, Andrew <andrew at donehue.net> wrote:

> I have spent a fair bit of time on google trying to find a solution for this
> problem, but no luck as yet.

Well, I am not an expert network engineer by any stretch of the
imagination, but I am curious to know how my understanding maps to
what experts say, and not embarrased to be wrong in public:

> For traffic management purposes I am trying to track packets based on
> incoming and outgoing interfaces (in a multi-homed network, presume 100%
> linux minus the switches).   I want to achieve this to allow exceptions for
> lower cost data movement (eg, if bandwidth is going via peering instead of
> transit then allow it to go through at a faster rate).  This is not to due
> to a per-GB download charge (just to the fact that the peering interfaces
> are faster than the transit) - so it is to avoid congestion.

...so, basically, you want the flow performance to match the bandwidth
of the links it will eventually take?

That sounds like a job for the standard flow control mechanisms in TCP
and similar protocols, which should start to feed back congestion
information from the busy point and automatically regulate speed from
the sender.

If you want to get fancy, implementing some better feedback mechanisms
(some versions of RED are, I understand, very effective in the role)
that start to slow traffic before they hit the congestion bottleneck
will help, but basic management of your buffers so the congestion
point doesn't introduce too much latency should do exactly what you
want.

Which makes me think that I have misunderstood your goals, or problem, here?

Regards,
    Daniel
-- 
✉ Daniel Pittman <daniel at rimspace.net>
⌨ daniel at rimspace.net (XMPP)
☎ +1 503 893 2285
♻ made with 100 percent post-consumer electrons