dkvz / many-threads-experiment

Experiments with the performance impact of having many threads (that aren't doing much) in various scenarios

Experiments with threads and performances

Experiment ideas

• A lot of sleeping threads that are set with a long duration;
• Sleeping threads with lower duration, that immediately go back to sleep;
• Threads that call "yield_now" immediately if a condition isn't met -> They didn't lie in the doc this uses a lot of CPU which makes sense since there's no blocking at all;
• Threads waiting for a Condvar - This should have the least impact.

Sleeper threads

• These generate a context switch every time they're wait period expires, or somewhere close to that which, I believe, is what's generating the load.
• From Process Explorer we can see these are in state "Wait:DelayExecution" which is not the usual waiting thread state.
• The state we want is called "Wait:UserRequest".

-> Could probably achieve this using Condvar as mentioned above, a blocking message channel or thread parking.

Testing "protocol"

• Disable antimalware
• Stop unrequired services if I have a script for that ready on target machine
• Disable bluetooth
• Pause Windows Updates