On Mon, 2007-03-19 at 14:41 +0100, Klaus Stock wrote:
> > For example. Let's say I've a DSL connection, 1Mb download, 512
> > upload. What I ideally would want to have is QoS for VoIP (I've a PBX
> > in my network). So I want VoIP to always have priority in the network.
> > But I do not want to "waste" bandwidth. For what I understand, the
> > "pipe" is always reserved, and if VoIP is not being used, other
> > traffic will not be able to use all the available bandwidth.
> > Sorry if the explanation is a little confusing... What I mean is: Is
> > for example my p2p client able to use all the available bandwidth (if
> > no VoIP connection is alive), and in the moment that a VoIP call is
> > being started "prioritize" the VoIP packets, so they arrive on time?
> The pipe is not being reserved in the way you mean it. At least not in a
> typical set-up, as the one whihc is produced by the "magic shaper wirzard".
> You use queues for the priorization of the traffic. The traffic is then
> allowed to use the pipe accoridng to it's priority. If no other traffic
> occurs, P2P trffic gets the whole bandwidth of the pipe.
> However, the pipe needs to be configured in a way that the buffer of the DSL
> modem never gets filled. Because as soon as packets get to the modem, they
> are out of reach of the m0n0wall. And the modem will sequentially process
> it's buffer, regardless of any QoS demands. For that reason, the pipes are
> typically configured with 10% less bandwidth that the "native" bandwidth of
> the modem/DSL line.
> Note that you'll most likely need NO traffic shaping for the downstream,
> since your LAN has most likely significantly more bandwidth than your DSL
> downlink. As there's no danger of some network buffer filling up, you can
> set the corresponding pipe to a bandwidth value which exceeds the DSL
> downstream bandwidth. No waste of bandwidth there!
> On the upstream however, you'll need the loss of 10% of the bandwidth. In
> return you receive a *much* better latency!
> Best regards, Klaus
The idea of QoS is that high priority packets are plucked from the queue
ahead of lower priority packets.
The worst case comes when the queue holds only big low priority packets
and just after one is picked, a high priority packet enters the queue.
This shows the value of smaller packets for multi-priority flows. ( for
example, 53byte cells (packets) used in the ATM protocol).
Best regards Bob G