High-end mass storage systems are becoming increasingly popular as
fabric elements in supercomputing facilities where distributed data
analysis is carried out. Clients of these Grid services are often
faced with slow-downs in the last mile of their computation jobs
which require data processing and visualization at their local
desktops. In addition, response latency is highly varied for the
high-end storage systems, because of contention for the same shared
resource. These huge data repositories also represent central
points of failure. In LAN environments present at these facilities,
individual desktops represent nodes with high speed network
connectivity and large potential unused disk storage.
        The Freeloader project is an effort to aggregate unused disk
storage from these desktops to form a large shared cache/scratch
space for large immutable data sets. Disk space is donated on a
voluntary basis by the user of the desktop machine. The collective
I/O bandwidth from these desktops  promises to provide performance
that is comparable to that of high-end storage access.
Collectively, these individual desktops provide services such as
reliability, high performance, availability and load balancing. In
this talk, we will be presenting the Freeloader architecture and
an initial proof-of-concept prototype. We discuss Freeloader's
performance in terms of access rates when compared with GridFTP
access to the GPFS parallel file system, and HPSS mass storage
access. We also discuss the impact of this resource borrowing on
each individual desktop making a contribution.