Many (most?) DSL and cable modem users have a 128 kbps uplink (for example, 1500/128), which is 16000 bytes/second. Typical MTU is 1500 bytes, which would take 94 ms to transmit over a 128 kbps uplink.
94 ms is a *LONG* time for VoIP. It seems inevitable that voice quality will be degraded unless MTU is decreased or uplink bandwidth is increased. I decreased my MTU from 1500 to 576, which is the lowest MTU value allowed by the DD-WRT firmware for my Linksys WRT-54G. That would reduce my maximum transmission (serialization) delay from 94 ms to 36 ms, which seems a lot more likely to be a delay that the jitter buffer could handle. Doing this reduces my effective bandwidth for non-VoIP applications by 4% (since each 1500-byte packet would be split into three 500-byte packets, requiring 20 extra bytes for IP headers for each of the two new packet fragments). A 4% bandwidth penalty is no problem if voice quality improves.
http://www.voiptroubleshooter.com/problems/adsl.html http://www.voiptroubleshooter.com/problems/mtu.html
Unresolved questions:
1) Is this idea correct? Does everyone with less than a 384 kbps uplink need to reduce MTU to avoid voice quality problems?
2) Does the QoS function in the Sipura 2100 or the Sveasoft or DD-WRT firmware for WRT-54G automatically reduce MTU if it sees a 128 kbps uplink, or does MTU need to be manually reduced?
3) Why doesn't Sipura have an MTU setting on the Sipura 2100 (since it includes a router, and reduced MTU seems to be essential for QoS)?
4) Has anyone managed to get excellent voice quality on a 128 kbps uplink with heavy concurrent non-VoIP network usage? If so, how did you do it?
http://en.wikipedia.org/wiki/MTU_%28networking%29 http://www.packetizer.com/voip/diagn.../bandcalc.html http://www.telchemy.com/conferences/2005/ISPCON2005.pdf http://www.erlang.com/protocols.html