dummynet dropping too many packets

[rihad]

Eugene Grosbein wrote:
> On Mon, Oct 05, 2009 at 04:50:10PM +0500, rihad wrote:
> 
>>>> Where has TCP slow-start gone? My router box 
>>>> isn't some application proxy that starts downloading at full 100 mbit/s 
>>>> thus quickly filling client's 1 mbit/s link. It's just a router.
>>> While there is no or little competition for bandwidth from the router
>>> to clients, TCP would work just fine. I suspect your shaping policy
>>> makes heavy competition between clients. In this case, TCP behaves
>>> not-so-well without help of router's good shaping algorythms
>>> and taildrop is not good one.
>>>
>> Nothing fancy (i.e. no competition). Only tons of per-user pipes 
>> simulating the given throughput.
> 
> You've mentioned previously: "The pipes are fine, each normally having
> 100-120 concurrent consumers (i.e. active users)."
> This IS competition between TCP flows inside each pipe.
> 
Well, each user gets instantiated with a new copy of the pipe. Each such 
user counts towards the limit imposed by hash_size*max_chain_len for 
that pipe only. It would have been competition had I used dst-ip dst-ip 
0xffffff00 or similar and not dst-ip 0xffffffff, _then_ all 256 users 
(determined by the mask) would compete for the pipe's bandwidth. So the 
only competition is in the uplink at our main Cisco, I guess.