,
5 tweets,
2 min read
Read on Twitter
So I based my 'rdpscan' tool on that patch to 'rdesktop'. The rdesktop code is pretty good, but OMG does it love global variables. Therefore, to simulateously scan for multiple targets, I can't use either 'threads' or 'async', but must instead 'fork' copies of the program.
Such 'forking' was the original design of Apache and a lot of network programming, but my decades of experience has consisted of avoiding it, because it sucks. Now I'm trying to optimize it for the most scalability possible.
My first version of rdpscan defaulted to 100 workers, but now you can set the number of worker processes with --workers configuration parameter. I set it to 10000, but am getting only around 1700 because performance. 
Well, at least it's ~15 times faster than it was before. Now I need to figure out more optimizations. I think one issue is that I'm doing fork/exec, when I really only need to do fork.
Gah, why are my command-line screenshot having crappy colors? Maybe if I change opacity? Testing..
