Applying Scalable Distributed Data Structures to Tuple Spaces

1 Distributed Tuple Spaces Tuple spaces have been conceived [3] and extended under the influence of databases and messaging systems. However, there is no notion of data schemes, and queries are specified by templates, that may wildcard arbitrary fields of a tuple. Therefore, indexing tuples is a highly non trivial task, which is complicated even more by the advent of field hierarchies. As a result, the need of multi-dimensional indexing and range queries becomes apparent. The transformation of an arbitrary field to an interval, as introduced in [5], induces the transformation of a tuple to a hypercube. Since templates are syntactically identical with tuples, this transformation applies to templates, too. Based on this abstract representation, a necessary condition for matching is deduced, i.e. the inclusion of a tuple's hypercube within the template's hypercube. Furthermore, tuples have to be distributed among several tuple space servers, in order to achieve a scalable tuple space. Therefore, each server is assigned a hypercube, which induces its tuple domain, i.e. the set of tuples with hypercubes that have a non null intersection with the server's hypercube. If a tuple is written to the tuple space, it is stored on an arbitrary server, the tuple domain of which including the tuple. Then, queries may be confined to the subset of servers with tuple domains that include the template.