Over the past ten years, measurements of Internet behavior have been
made for a range of applications:  from research focused on
understanding Internet behavior in the form of invariant
characteristics to operational decision-making in Internet-based
companies dependent on the real-time state of the network.  However,
making representative measurements of any type in a large , dynamic,
heterogeneous environment such as the Internet is difficult.  This
paper outlines a number of problems facing researchers and companies
who want to use Internet measurement data in their studies or
operations.  It also describes the architecture for a Global Internet
Measurement Infrastructure (GIMI), which is significantly more
comprehensive than current measurement platforms.  GIMI is envisioned
to provide an application programming interface enabling researchers
to extract securely any type of Internet data from the infrastructure.
Its design and deployment will insure that data collected from the
infrastructure is inherently representative of a wide variety of
Internet behavior.  Significant hurdles, from determining where to
place measurement systems to how to correlate results across network
levels, must be overcome.  We describe these difficulties and report
on progress on basic aspects of the architecture made to date.
Finally, we argue that an infrastructure such as GIMI is necessary for
the next generation of characterizations and models of Internet
behavior as well as being an enabling technology for a variety of
real-time applications such as wide-area security and grid computing.