Jug is a task-based parallelism framework. Jug allows you to write code that is broken up into tasks and run different tasks on different processors. It uses the filesystem to communicate between processes and works correctly over NFS, so you can coordinate processes on different machines. Jug is a pure Python implementation and should work on any platform that can run Python.
libjmmcg is a basic, low-level library with pretensions to implementing features above and beyond (but not necessarily better than!) those implemented within the Standard C++ Library and the Boost Library. It features a library for multi-core or multi-chip SMP parallelism, a suite of hashing algorithms, functions for raising numbers to integer powers, a generic factory wrapper and a generic, multi-threaded, read-only cache (which uses PPD), arguably the world's worst sorting technique, trace output, exceptions that have the file, line, revision, function, and argument details, string utilities, logging, simple command line processing, and much more.
FastFlow is a pattern-based programming framework targeting streaming applications. It implements pipeline, farm, divide and conquer, and their composition, as well as generic streaming networks. It is specifically designed to support the development and the seamless porting of existing applications on multi-core. The layered template-based C++ design ensures flexibility and extendibility. Its lock-free/fence-free run-time support minimizes cache invalidation traffic and enforces the development of high-performance (high-throughput, low-latency) scalable applications. It has been proven faster than TBB, OpenMP, and Cilk on several micro-benchmarcks and real-world applications, especially when dealing with fine-grained parallelism and high-throughput applications.
YML is a research project that aims to provide tools for using global computing middleware such as GRID, peer to peer, and metacomputing environments. The YML software architecture enables the definition of parallel applications, independently of the underlying middleware used. Parallel applications are defined using a workflow language called YvetteML.
Charm++ is a portable adaptive runtime system for parallel applications. Application developers create an object-based decomposition of the problem of interest, and the runtime system manages issues of communication, mapping, load balancing, fault tolerance, and more. Sequential code implementing the methods of these parallel objects is written in C++. Calls to libraries in C++, C, and Fortran are common and straightforward. Charm++ is portable across individual workstations, clusters, accelerators (Cell SPEs and GPUs), and supercomputers such as those sold by IBM (Blue Gene, POWER) and Cray (XT3/4/5/6). Applications based on Charm++ are used on at least 5 of the 20 most powerful computers in the world.
The DUP System is a language for productive, parallel, and distributed stream processing on POSIX systems. Programming with DUP is similar to writing shell scripts with pipes except that filters can have multiple inputs and outputs. Furthermore, the computation can be spread across multiple computers. A distinguishing characteristic of DUP compared to other streaming languages is that filters can be written in almost any programming language. The DUP System distribution includes the runtime system and a collection of over a dozen multi-stream filters.
libpersist is a small library of C++ base classes and templates providing the capability to store and load objects in their entirety across executions. It also provides for object relocation between threads in a NUMA (non-uniform memory access) parallel processing environment. Objects are stored to disk as POD. They are reconstructed when reloaded into memory. Relocation is a special case of persistence, requiring reconstruction after the POD has been moved in memory. After all objects have been relocated, all references to the objects are swizzled (their IDs are resolved to their addresses), so that they can be dereferenced efficiently.