Experiments on Packet Dropper on Kernel 2.4.10 shows the performance drops significantly even when the dropping rate is 0.1%
Experiments on Nov 3 shows that the significant performance loss is because of the inefficient use of the "get_random_bytes" in kernel space
Experiments on Nov 5, using new version of packet dropper, packet is dropped one out of every 1/rate packets regularly.
Experiments on Nov 23, using new version of packet dropper, packets are dropped randomly. The result is totally different from the result from dropping packets regularly. (the efficient random number comes from CPU cycle and bit shift operaiont. )
Experiments on Dec 3, do the experiments almost automatically, count TCP packets dropped via /proc file system. New version of random number generation.
Experiments on Dec 3,There is significant burst and large delay happens when we regularly drop one packet out of every 6 packets.
Experiments on Dec 6, Found that using the rate*(1-rate)*MAXUNSIGNEDINT as threshold value is a mistake. Do experiments with rate*MAXUNSIGNEDINT
Still have difficulties in finding a way to implement packet delay for a target connection. . problems
My intuition: The one way delay of router ( doublin ) is about 145 uSec. See the calculation here
Experiments on the one way delay shows that the actual delay by the line and router should be close to 72 uSec, See experiment here