Future high-performance virtual machines will improve performance through s
ophisticated online feedback-directed optimizations. This paper presents th
e architecture of the Jalapeno Adaptive Optimisation System, a system to su
pport leading-edge virtual machine technology and enable ongoing research o
n online feedback-directed optimizations. We describe the extensible system
architecture, based on a federation of threads with asynchronous communica
tion. We present an implementation of the general architecture that support
s adaptive multi-level optimization based purely on statistical sampling. W
e empirically demonstrate that this profiling technique has low overhead an
d can improve startup and steady-state performance, even without the presen
ce of online feedback-directed optimizations. The paper also describes and
evaluates an online feedback-directed inlining optimization based on statis
tical edge sampling. The system is written completely in Java, applying the
described techniques not only to application code and standard libraries,
but also to the virtual machine itself.