Hi list, When running OpenBSD on QEMU without KVM, I see an increase in the number of sched_yield() system calls in certain multi-threaded benchmarks (sysbench). However while using KVM accelerator, the number of this system calls is minimal. Does any have any insight why this is happening? The execution of the benchmark is really slowed when not using KVM which is to be expected but why is there such as massive increase in sched_yield system calls? Moreover when using icount the number of sched_yield calls is further increased.
Some Number: With KVM accelerator - #of sched_yields = 500 Using TCG - #of sched_yields = 2600 Using icount - #of sched_yields = 16000 Anyone has an idea why this might be happening? Thank you!
