Distributed memory in a heterogeneous network, as used in the CERN-PS complex timing system

The Distributed Table Manager (DTM) is a fast and efficient utility for distributing named binary data structures called Tables, of arbitrary size and structure, around a heterogeneous network of computers to a set of registered clients. The Tables are transmitted over a UDP network between DTM servers in network format, where the servers perform the conversions to and from host format for local clients. The servers provide clients with synchronization mechanisms, a choice of network data flows, and table options such as keeping table disc copies, shared memory or heap memory table allocation, table read/write permissions, and table subnet broadcasting. DTM has been designed to be easily maintainable, and to automatically recover from the type of errors typically encountered in a large control system network. The DTM system is based on a three level server daemon hierarchy, in which an inter daemon protocol handles network failures, and incorporates recovery procedures which will guarantee table consistency when communications are re-established. These protocols are implemented over a communications layer which performs the data conversions, packet splitting, error-correction with retry, and time out handling. The same communications layer is used to implement the application program interface which calls on the server daemon for client services. DTM is a registration based system, in which communications are established dynamically as needed, and tables are distributed only to the clients who have registered their interest in them. The registration protocols include mechanisms to recover from daemon re-launches, and clean up after aborted clients.