For now, please try to bear with these:

• NotSoTinyWM

  A collection of minimal window manager feature examples. Learn more or see the code.
  
  Language: Implementations in both C and Python
  

• Per-process network monitor for linux

  It's like top for network!
  Currently slow as hell, and inaccurate when there are less than 1-2 packets per second.
  Good for finding out what process is causing network spikes.
  
  Language: Python
  Download: ppntop.py
  
  Update: Have a look at NetHogs if you need this functionality, it's way faster/nicer

• IO-trace (x86 in/out instructions) for linux

  Shows how a program is accessing hardware io. (Got a huge program that you know accesses a few io-ports, but don't feel like flipping through the whole source / don't have the source? No problem)
  Again, currently slow as hell. This single-steps the target program. If I have the time and motivation, I'll rewrite this to:
  • trace syscalls
  • prevent the target program from using iopl/ioperm
  • if the target receives SEGFAULT, check if the faulting instruction is in/out
    • if not, let it die
    • if it is
      • re-run the desired iopl/ioperm
      • re-run the current instruction
      • disable the iopl/ioperm again
      • continue execution "normally" (trace syscalls again)
  I think that should work, but last time i tried mixing SYSCALL and SYSEMU with ptrace, it just didn't work (but that was ages ago). Another possibility would be to just always use SYSCALL, and perform a no-op syscall when the syscall needs to be omitted (also not sure how ptrace likes when you substitute the current non-syscall instruction with a syscall).
  If you see anything else obviously wrong with this approach, please drop me a mail :)
  
  Language: C
  Download: iotrace.c