[Caught by the no-procrastination feature by accident.]
Our earlier (dettrace) prototype allowed syscall-free regions in separate processes to run physically in parallel. Hermit actually hasn't added any process parallelism yet, but it's designed to actually go further than dettrace in this respect.
Specifically, Hermit is architected so that the thread-local syscall handler "checks out" resources from the central scheduler. Resources include things like paths on the file system, contents of files, shared memory, and permission to perform external side effects. Right now, all requests wait for the scheduler to background all other threads and make the current thread the only runnable. But the idea is: in the future we will keep the semantical identical log of linear "commits" (linearization), but will simply background the current thread while it uses the resources they checked out, move forward to the next scheduler iteration, and only block the next runnable thread until its requested resources are freed, not until ALL other threads have finished their timeslice and gone back to waiting on the scheduler.
Our earlier (dettrace) prototype allowed syscall-free regions in separate processes to run physically in parallel. Hermit actually hasn't added any process parallelism yet, but it's designed to actually go further than dettrace in this respect.
Specifically, Hermit is architected so that the thread-local syscall handler "checks out" resources from the central scheduler. Resources include things like paths on the file system, contents of files, shared memory, and permission to perform external side effects. Right now, all requests wait for the scheduler to background all other threads and make the current thread the only runnable. But the idea is: in the future we will keep the semantical identical log of linear "commits" (linearization), but will simply background the current thread while it uses the resources they checked out, move forward to the next scheduler iteration, and only block the next runnable thread until its requested resources are freed, not until ALL other threads have finished their timeslice and gone back to waiting on the scheduler.