M3 is a new hardware/operating system co-design that targets very
heterogeneous platforms, containing mixed ISAs and accelerators. M3 uses
a microkernel design that places components onto different cores and
uses a dedicated hardware component for message-based communication
between these components.
Currently, most APIs in M3 are synchronous. That is, after component A
sent a message to component B, component A idles until the reply from
component B arrives. Since M3 prefers to place components onto different
cores (having both performance and security advantages), this
potentially wastes time, because component A's core cannot be used for
other work during that time.
M3 provides user-level threads to tackle some shortcomings of the
synchronous API. In this thesis, various approaches to an asynchronous
API in C++ are evaluated regarding their usability and their
performance. Precisely, callbacks, event-handling state machines and
coroutines are implemented and compared to the existing user-level threads.
