on_report() is always triggered per ACK about portus HOT 4 CLOSED

Hi, thanks for this report. Can you try the following steps:

1. ccp_kernel_unload (as root)
2. run this const algorithm: python3 <..>/portus/python/example/const.py
3. ccp_kernel_load ipc=0 (as root)
4. now start e.g. iperf sender and receiver with something like iperf -s -p 5001 and mm-delay 25 mm-link --uplink-queue-args="packets=200" --uplink-queue="droptail" <..>/bw24.mahi <..>/bw24.mahi -- sh -c "iperf -c \$MAHIMAHI_BASE -p 4242 -Z ccp -t 30 -i 1"

If those steps do not work, it would be helpful if you provided the outputs in /var/log/syslog between steps 2 and 4.

from portus.

Everything works fine after reloading the ccp kernel! Thanks for your help @akshayknarayan!
I would be really appreciated if you could give some hints about why it works :)

from portus.

My guess is that when running this modified const after various other tweaked versions, the datapath program table ran out of space (https://github.com/ccp-project/libccp/blob/master/ccp.c#L319). In ccp-kernel we set the size of this datapath program array to be 10 (https://github.com/ccp-project/ccp-kernel/blob/master/tcp_ccp.h#L11). Because the report struct is the same size as the previous program, it's harder to notice this (without looking for the log in /var/log/syslog), since the previous datapath program will run (in your case, the one that reported per-ack) and send the same report. Unloading and re-loading the kernel module of course frees and reallocates this array.

This is a bug - ideally we would have some mechanism for removing old datapath program entries once there is no more space. I will try to fix it when I have time.

from portus.

That makes sense, thanks for your detailed explaination!

from portus.