[Bloat] Bufferbloat

[Dave Täht]

This is feedback from the Director of Technology at a large school
(private high school) system regarding figuring out bufferbloat in his
environment. 

Any thoughts?

>>> When I get some time I would like to look a little closer to see if
>>> this affects us in a significant way, but right now it doesn't seem
>>> broken (translation: nobody is complaining) so I have other things
>>> that demand my time
>>> more
.
>>It's an interesting puzzle. Most people are reporting enormous gains in
>>voip, dhcp, ntp, DNS, and other udp based protocols, once they fix
>>latency under load.
>
> Perhaps I didn't look hard enough, but most of what I read regarding
> testing and troubleshooting seemed to apply to very small
> networks. While I could test individual clients I don't think this
> would be terribly useful and worth the time. What I really need to do
> is find a way to get stats on the performance of our entire network's
> Internet connectivity during peak times to see if latency is a
> problem. During these times we easily have 700+ clients and thousands
> of flows and I have no idea how to get aggregate stats that are useful
> to diagnose bufferbloat on our network. Of course we also have QoS (we
> use the NetEqualizer which is a great little box)
> which intentionality delays packets of our heaviest users so I would
> have to find a way to factor effects due to the QoS out of any
> stats--turning off the QoS during peak times is not an option. If you
> know of pages which I missed with discussion about mid sized networks
> like ours please send the URL's my way.
>
> Take care Dave and ping me if you're ever in the area.

-- 
Dave Taht
http://nex-6.taht.net