208 High Performance Parallel I/O
put() operator, and the visualization processes can asynchronously fetch rel-
evant portions of the data using the corresponding get() operator. Note that
these operators are independent of the scale and the distribution of these
interacting applications. The internal data management mechanisms in Data-
Spaces ensure the scalability of the distributed data storage and lookup mech-
anisms. The DataSpaces runtime builds on DART [5], which enables direct
low-overhead, high-throughput memory-to-memory communication between
the interacting nodes using Remote Direct Memory Access (RDMA).
DataSpaces implements a distributed query engine with simple and flexi-
ble query semantics to facilitate access to the data. An application ...