Title: A Framework for Reducing the Cost of Instrumented Code
Authors: Matthew Arnold and Barbara G. Ryder
Institution: Rutgers University

Abstract:

Instrumenting code to collect profiling information can cause
substantial execution overhead.  This overhead makes instrumentation
difficult to perform at runtime, often preventing many known 
offline feedback-directed optimizations from being used in online
systems.  We present a general framework for instrumenting code that
uses fine grained sampling to allow previously expensive
instrumentation to be performed accuratly with low overhead.  Our
framework does not rely on any hardware or operating system support
and is fully tunable; the sample rate can be adjusted at any time to
match the type of instrumentation being performed.  By reducing the
overhead of instrumentation, our framework eliminates one of the
biggest obstacles to performing feedback-directed optimizations at
runtime.  We present experimental results validating the overhead and
accuracy of our technique.