Speedup Tsung starting when using hundreds of clients about tsung

[tisba]

Wow, this is awesome! A while back I was thinking about how to improve this without making to much changes, but this approach seems to be simple and straight forward. Thanks!

[nniclausse]

i was thinking about your use case , when i realize that it was silly to not spread the list of beams, and by doing this we could go much higher that the 10 parallel start.
I did a test with 620 beams (on ~50 nodes), it tooks 1mn28s (and most of the time is spent in the global sync phase, so there is still room to optimize it further)

[tisba]

I started to work on researching what is necessary to use -connect_all false in tsung and only establish explicit connections between controllers and slaves and not between slaves. It's a bit tricky, because global name registration does not work in that case, but that should not technical not be necessary. It does require a bit of rework though.

If we can get this done, there will only be one connection per slave, and not (slave^2-slave)/2 connections.