Not sure how you maintaining QEMU makes you a credible source for evaluating a schedulers performance. It's apparent to me the performance of the scheduler is a function of the workload, so YMMV.
I worked on a project where we collected detailed production runtime characteristics and evaluated scheduler algorithms against it. Tiny improvements made for massive savings.
I definitely correct my "does" a good job with "did" a job. But ultimately, I've advised a good deal of caution, which I think is fair, in particular, considering that only a small fraction of the companies has a compute scale where tiny improvements make massive savings.
Comments
Not sure how you maintaining QEMU makes you a credible source for evaluating a schedulers performance. It's apparent to me the performance of the scheduler is a function of the workload, so YMMV.
I worked on a project where we collected detailed production runtime characteristics and evaluated scheduler algorithms against it. Tiny improvements made for massive savings.
I definitely correct my "does" a good job with "did" a job. But ultimately, I've advised a good deal of caution, which I think is fair, in particular, considering that only a small fraction of the companies has a compute scale where tiny improvements make massive savings.