[altq 310] Re: Ingress traffic shaping?

[George Uhl <uhl@mamba-e.gsfc.nasa.gov>]

> Unfortunately the config yields 61.83K/sec on Input traffic (i.e. no
> discernible shaping).
> 
> Changing the root interface speed from 1024kbs to 961kbs (roughly 90%) did
> nothing for the 'incoming' data rate (OK, it made it go from 61.83K/sec to
> 61.78K/sec - not much!)

Just curious, did you try shaping the input traffic using the ALTQ
traffic conditioner?  It allows you to drop incoming packets
according to some rules.  It is not a queuing discipline like
CBQ so you can't borrow bw if it's available.  I notice you
haven't permitted borrowing on either CBQ class anyway, so
why not try using the token bucket meter as a traffic shaper?


> Downside is that I have to run two instances of probes/monitors etc. to watch
> things, but it works...
> 

Yeah, that is somewhat of a nuisance.

> I've not had time to actually look at the traffic going through this system -
> I will if I get more time...
> 
> Thanks to all who suggested fixes / ideas,
> 
> My final question is - can anyone see anything "bad" with this set-up? i.e.
> 'very' inefficient? Likely to cause problems etc?
> 

Not really.  MRTG would be useful for long-term monitoring of the link.
It requires some hacking of cbqprobe and cbqmonitor but it could
payoff.  I've done it and I ought to clean it up and make it
available to Kenjiro (along with some other stuff).  I've set it up so
I can select and observe per-class throughput in bps and packet count on
any CBQ-enable i/f. Would you find something like this useful?

With MRTG I noted periodic outages of ALTQ CBQ forwarding service
on a 30Mbps TCP flow over a 24 hour period that K.C's latest patch is 
supposed to fix.

> The machine running this is quite fast (P2-450), with three fxp's - so
> shouldn't be a problem? :)

No, especially at the very low traffic rates you are dealing with.
I have a mix of PCs.  One is an old P.Pro 200 with 3 x OC-3
ATM cards configured as a Diffserv interior router.  I haven't 
observed any performance degradation due to traffic load on it yet.


George Uhl