htb issue with large q limits

 Below is a perplexing problem that I am hoping the LARTC kernel
experts can help me with.


We have a traffic control system based upon htb which when packets are
directed to it from a bridge between eth2 and eth3 works great.
Accurate adherence to specified link rates based upon priority is
achieved.

For example, if I have a 2 Mbps UDP stream going to Q-PRI4 and another
2 Mbps UDP stream going to Q-PRI5 with a link rate of 800Kbps, then
800K of traffic is sent from Q-PRI4 with and the rest of the packets
on Q-PRI4 are dropped, and 100% of the packets directed to Q-PRI5 are
dropped.  This is exactly how it should work.

BUT using all the same traffic control rules, if we intercept packets
on eht2 with PF-RING and then output to eth3 with raw socket packet
sends , the QDisc does not work the same.  Instead Q-PRI4 and Q-PRI
both get 400K of traffic.  The higher priority of Q-PRI4 is not
respected.

Why does it behave differently when intercepting packets from PF-RING
and then doing raw sends  versus how it works with a classic bridge ?

Another observation is that if I have a queue limit of 45 or less.
Then it works correctly in both scenarios.  But if I have a queue
limit higher than 45 then it disregards the priority of the Q.