may add loss punish when one path get loss

[fareign]

in the picture. wan1 is home fiber broadband ,but QOS by ISP，only can run 3mbps. wan3 is 5G simcard. singnal is not good,but can run 4-6mbps without loss. I not sure why dispath suddenly high ,when wan1 > 3mbps ,so it must with lot of loss.

so facing this situation,how to do?

[mp0rta]

Could you clarify what "dispatch" means here? Is it a metric shown in the OpenMPTCProuter?
Also, I'm not currently involved in the integration with OpenMPTCProuter, so I'm not sure if I can give an accurate answer.
BTW, mqvpn is still a product under development (at least until version 1.x),
and because it's a personal project, real-world testing hasn't kept up yet. Issue reports are always welcome.

[fareign]

yes,is openmptcprouter 。
dispath mean dispense? I mean is , if wan1 send bandwidth greater than 3mbps , may couse some loss. and at this time WLB send much more packetto compensate the loss packet.more packet send and more packet loss，it may lead ot endless loop.
so,this moment,we shoud reduce wan1 and increase wan3 may a good select?
I apologize for my unclear expression，thanks!

[mp0rta]

I'm not entirely sure I understand the situation, but is your concern the following?
wan1 is limited to 3Mbps by ISP QoS, and packets are dropped when it exceeds that limit.
When loss occurs, the total amount of transmitted packets increases, causing even more traffic to flow through wan1, which leads to further loss — a vicious cycle.

mqvpn uses QUIC DATAGRAM frames for tunnel traffic, which are no retransmission. So loss on wan1 maybe not directly cause mqvpn to increase the transmission volume. However, WLB does have a scheduler that takes loss rate into account — if loss on wan1 exceeds a certain threshold, WLB evicts flows from wan1 and shifts them to wan3, so it should not fall into a vicious cycle unless bugs.

That said, ISP QoS bandwidth caps are a special case. Congestion control periodically probes for available bandwidth, so oscillation from this probing could repeat.

I'm curious about the actual loss rate. If you're able to provide a graph of the overall loss rate, that would be helpful.