Task Scheduling

I have come to revisit the low priority issue with task scheduling. I mentioned years ago I tested the program on an EPYC with 64 cores/128 threads. I managed to only reach 22, which is pretty low. I will therefore revise the scheduling process for the thread pool, somehow making it faster. I do have some ideas, like multiple queues(Used in v1), task ownership and then some. It is still in the early stages of research before I start to write some sort of implementation where I can compare each solution to see which one will reach 128 active threads first. And if all, which one will have the least overhead and will scale better for different type of tasks.

This will be an experiment, planned to be connected to my university as my first real paper. I see it as an continuation to my previous work over thread pools, improving the current implementation significantly. Maybe even make it possible for it to scale over a whole cluster, if needed. Sure, it is not so useful for PixelMap, as the biggest worlds takes only 15 minutes to process and generate an image. The problem is if the plan to later create smaller images for a genuine browser map, or even create isometric images, then the additional parallel computation may come in handy. Also, reducing overhead minimizes unnecessary computation.