[LARTC] hfsc rule command problem

[Andy Furniss]

Thossapron Apinyapanha wrote:

> tc class add dev eth2 parent 1: classid 1:1 hfsc sc m2 10000kbit 

If it's really 10mbit eth, 10mbit rate is too high.


> tc class add dev eth2 parent 1:11 classid 1:199 hfsc rt m1 3500kbit d 10s m2 500kbit ls m2 3500kbit ul m2 3500kbit
> tc qdisc add dev eth2 handle 199: parent 1:199 sfq

sfq is really for bulk, with the added complication that it won't quite 
be right on an hfsc rt class - if backlogged hfsc rt de/re queues to get 
the next packet length, but sfq (as I read it) doesn't reset the "turn 
pointer" when it requeues. Probably OK if you really need it I suppose - 
but then rt classes should probably not get backlogged (excepting bursts 
  maybe)


> Real time class         guarantee rate: 3500kbit        max rate: 10000kbit                
> VoIP                         guarantee rate: 200kbit          max rate: 3500kbit
> MMS                         guarantee rate: 1300kbit        max rate: 3500kbit
> Telnet                      guarantee rate: 1500kbit        max rate: 3500kbit 
> Default                    guarantee rate: 500kbit          max rate: 3500kbit

They are not going to be rt as such once over their guaranteed rates.

> 2.       Is it true about HFSC can’t manage traffic load more than 1Mbit ??

The problem is I think, your use of default - If you don't use it hfsc 
will drop unclassified traffic, htb will let it through unshaped, so it 
is in someways needed more on hfsc (though you can filter non ip).

You could use it but not send any ip traffic to it, or you could 
explicitly filter arp (& other non ip) to a dedicated rt class with 
enough bandwidth. Delaying/dropping arp is not something you should do...

Andy.