In any multiuser distributed computing environment, there is
usually a set of files which are accessed frequently by the user community and
a set of disks connected via a local area network (LAN), used to store these
files.  It is the system administrator's job to decide which files are to be
stored on each particular disk. Usually the disks available for storage are
physically separated and connected via a network. A Network File System (NFS),
handles all the file access requests. The disks typically have different
characteristics since distributed networks usually have heterogeneous
resources. The placement of heavily accessed files, across these disks, can
have a large impact on the performance of the system in terms of the
throughput. An effective file placement strategy should consider file access
probabilities and disk characteristics. Placing all the files on the fastest
available disk may not be optimal due to queuing and contention delays.
Typical results of a poor file placement include low throughput and 
overloading of ``hot'' disks. Modeling a distributed system is not trivial, 
since it includes modeling the distributed computing resources, the network 
connecting them to the file servers, and the overhead introduced by the NFS. 
This paper presents a simple queuing network model of a distributed system 
and a method to efficiently calibrate the model with experiments. Certain 
approximations are made to simplify the analysis and calibration. Validation 
experiments comparing the predicted performance with the actual performance 
justify the approximations and indicate that simple models can serve the 
purpose of accurately indicating potential improvements in performance.