Developing a Lock-Free Scheduling on Manycore Computer

Increasing scale and the need for quick response to heterogeneity is hard to meet with current monolithic manycore scheduler architectures. This limits the rate at which new features can be made effective use of, decreasing efficiency and utilization, and eventually restricting manycore systems' growth. We propose a new approach to meet these needs by parallelism, shared resource state, and transactional memory mechanism, which is a lock-free optimistic concurrency control.We compare this approach to existing manycore scheduler designs, assess how much interference occurs between schedulers and how much it matters in practice, suggest some techniques to alleviate it, and finally discuss the case highlighting our lock-free approach - all driven by current popular I/O intensive and CPU intensive workloads.