Hello!
> i observe the same results regardless of the `tc qdisc ... sfq ...'
> for every class.
sfq works inside class, so that it cannot affect
inter-class sharing.
> and the traffic goes to the correct class, but using netpipe as a
> simple throughput test i'm getting:
It looks almost OK. The deviations may be explained by
senders irregularity. Do you really run all the test simultaneously?
Also, did you try to use builtin estimator, in this case it
may provide better estmate. BTW did you monitor backlog sizes?
> what i dont't understand is that:
> tc -s class show dev eth0
>
> class cbq 1:1 parent 1: rate 8Mbit (bounded,isolated) prio 1
> Sent 183336411 bytes 129013 pkts (dropped 0, overlimits 0)
> rate 191bps 2pps
> borrowed 46807 overactions 0 avgidle 3956 undertime 0
>
> tells me that 1:1 borrows some traffic, although it's isolated and
> bounded (i think it shouldn't borrow frm it's ancestores and its
> neighbours).
On nodes classes "borrowed" means borrowed from it 8)
Well, I see it is confusing, but it is still correct
from grammar veiwpoint: only add "from this class" to nodes
and "by this class" to leafs. 8)
Alexey
-
To unsubscribe from this list: send the line "unsubscribe linux-net" in
the body of a message to [EMAIL PROTECTED]
