Robert Reed discusses task scheduler partitioning with Intel TBB.
“I thought I was ready to dig into the scheduler using the tools I’ve built so far. But I was disturbed by some of the concurrency diagrams my current data collector and post-processing program generated. Here’s a typical case:
This diagram to the left appears to show a process where threads 1 and 2 are lazily working on one major index while thread 6 burns through ten of them, then helps thread 2 on index 25.Can this really be what is happening?Perhaps it’s time to revisit the details of the collection and analysis process and try to improve confidence I’m really collecting and representing the correct data and analysis.”
