I set up some QoS for VoIP. My provider told me that they're using G.711u so it should be using 64kbps payload rate (w/o overhead), so I figure 70kbps. If I set Tomato 1.02 to 68-70kbps on highest priority, set via IP address, I get choppy voice. If I set highest to 75-95kbps, the choppiness goes away. So, I got to thinking, could it be that the 5 queues (ignoring ABCDE) in tomato assume bytes/sec but fail to consider the VLAN tag overhead (4-bytes for 802.11p and 802.11q)? It is even more important for VoIP because of the small frames used (64B but I am not entirely sure; surely much much less than 1514B). Can anyone confirm this? If so, it might make more sense why I had to set the bits/sec so high. Also, if you turn off QoS, does that effectively not use 802.11p/802.11q (thus no extra 4-bytes overhead penalty)? I'm also not sure where the downstream QoS is considered ("shaped" via dropping ACKs), is it on the ingress WAN interface (w/o VLAN tag) or on the egress LAN interface (w/VLAN tag)? Similarly, not sure about upstream traffic, I assume it is shaped on the ingress LAN interface (so shouldn't the VLAN tag 4-byte overhead be considered)? If it shapes on the LAN interface (which has VLAN tag), then there might not be a way to correctly "reverse-engineer" the WAN shaping because the frame size wouldn't be known in advance of the shaping (if it truly shapes based on kbps). It is even more complex (if not impossible) if shaped based on packets/sec because of the vastly different sizes between max and min packet sizes.