Object Server and Simple Fast Serialisation
This packages provides two concepts: the ability to serve objects from an R process asynchronously using threads (no forking necessary) and a simple fast serialisation (SFS) format for transfer of objects between R processes.
Asynchronous, Threaded TCP Object Server
The server allows the R session to register R objects under a name (key) which can be retrieved via TCP ansynchronously, i.e., while the R session is performing other tasks. The main intended use is to provide in-memory "hot" storage of objects that can be both computed on and sent to other computing nodes on demand.
The start function starts a TCP server on a separate thread from the R process. It then responds to any queries to fetch or remove objects from this session's object cache.
put(key, value) registers the object value to be served under the name key. Because the server is asynchronous, any removals of objects are not performed immediately, but instead R can issue clean() command to perform any removal of R objects that were requested by the clients in the meantime.
In addition to the server, the package also provides a simple client which can request objects from another R process via TCP using the ask() function. It is more efficient and faster than using R's socket functions.
Simple Fast Serialisation
In order to avoid the necessity to store serialised form of objects which are already in the R session this package also introduces a Simple Fast Serialisation format. It is designed to be fast and streamable, i.e., it requires no memory buffer to serialise or unserialise (technically, it requires 8 bytes). The implementation supports direct (un-)serialisation to/from memory, file or socket. Currently, most but not all object types are supported.