January 7, 2023

BSD TCP/IP for Kyu - more testing

Here is the python script that I have been using on my Fedora linux system to test my TCP implementation: As you will notice, this runs netcat (which is called "ncat" on Fedora) to connect to port 114 and send the command "big" to the server on the Kyu system (named "levi"). This causes it to respond with 200,000 bytes (see test.c in the Kyu tcp directory).

I have now run this for over 48 hours. Recently I started "ping" while it was running and everthing worked just fine for over an hour. Now I am going to try running 3 separate instances of the "try_tcp" script and see how that goes. 

 Thread:       name (  &tp   )    state     sem       pc       sp     pri
  Thread:      shell (400a7a1c)   READY J          40010560 40568000   11
  Thread:     net-in (400a764c)   READY J          40010560 40570000   20
  Thread:    net-out (400a7558)     SEM J net-outq 40010560 40572000   21
  Thread:   net-slow (400a7464)  REPEAT C          40015e68 40574000   22
  Thread:  tcp-input (400a7834)     SEM J tcp-main 40010560 4056c000   24
  Thread:  tcp-timer (400a7740)  REPEAT C          400243ac 4056e000   25
  Thread: wangdoodle (400a727c)     SEM J   socket 40010560 40578000   31
  Thread:   wang_thr (400a7370)   READY J          40010560 40576000   35
* Thread:   wang_thr (400a7188)   READY I          4001d1d4 4057a000   35
  Thread:       idle (400a7928)   READY C          40010a80 4056a000 1234
It works OK, but some of the threads get starved (all the connection handling threads run at priority 35, and Kyu makes no guarantees about how it will handle competing threads at equal priorities.

When all this finishes, I see the Kyu output size claiming to have 2 entries. And when I look at this I catch a signficant oversight, there was no lock around the code where I place entries on the Kyu output queue. So I have fixed this and done a quick test.

Try all this on the BBB

There is no reason to think that tcp would no work on the BBB if it works on the Orange Pi. Actually I have been using a NanoPi Neo board for all of this development and testing. It has an Allwinner H3 chip, 512M of ram and is all but identical to the Orange Pi PC that I also use.

Have any comments? Questions? Drop me a line!

Kyu / [email protected]