Page migration problems arise in distributed data management. The goal
is to distribute a set of pages in a network of processors, each of
which has its local memory, so that a sequence of memory accesses can
be executed efficiently. Most previous work assumes that the local memories
have infinite capacities, which is unrealistic in practice. In this
paper we study the migration problem under the assumption that the local
memories have limited capacities. We assume that the
memories are {\it direct-mapped}, i.e., the processors use a hash function in
order to locate pages in their memory. We show that, for a number of
important network topologies, on-line algorithms with a constant
competitive ratio can be developed in this model. This is essentially
the first work on page migration that makes realistic assumptions
concerning memory capacity and develops upper bounds that are meaningful
in practice. )
%)
We also study distributed paging. We examine the {\it migration version} of
this problem in which there exists only one copy of each page. We develop