Automatic exploitation of data parallelism in streaming applications

A system comprising:

• a compiler executing on a processor, said compiler determines whether an operator of the streaming application is safe to parallelize based on a definition of the operator and an instance of the definition, the definition of the operator including a template of the operator, the instance of the definition including a modified version of the template, the operator being not safe to parallelize when at least one of:
• the operator has a non-partitioned state,
• the operator has selectivity greater than 1, wherein the selectivity is a number of output tuples generated for each input tuple,
• the operator has more than one input port,
• the operator has more than one output port, and
• the operator is depended on by another operator, wherein the dependency is a co-location constraint,
• said compiler determines that the operator is not safe to parallelize when the operator has non-partitioned state, said compiler forms parallel regions within the streaming application when the operator is safe to parallelize, said compiler determines synchronization strategies for the parallel regions, the synchronization strategies being determined based on the definition of the operator and the instance of the definition;
• and
• a runtime system connected to said compiler executing on said processor, said runtime system enforces the synchronization strategies of the parallel regions, said runtime system uses round-robin and sequence numbers and pulses as the synchronization strategy as instructed by said compiler;
• and
• performing a shuffle when two adjacent parallel regions include incompatible keys.

A computer program product for exploiting stateless and stateful data parallelism in a streaming application, said computer program product comprising a non-transitory computer readable storage medium having stored/encoded thereon:

• first program instructions executable by a compiler to cause the compiler to determine whether an operator of the streaming application is safe to parallelize based on a definition of the operator and an instance of the definition, the definition of the operator including a template of the operator, the instance of the definition including a modified version of the template;
• second program instructions executable by the compiler to cause the compiler to form parallel regions within the streaming application when the operator is safe to parallelize;
• third program instructions executable by the compiler to cause the compiler to determine synchronization strategies for the parallel regions based on the definition of the operator and the instance of the definition;
• fourth program instructions executable by the compiler to cause the compiler to determine that the operator is not safe to parallelize when the operator has selectivity greater than 1, the selectivity being a number of output tuples generated for each input tuple;
• and
• fifth program instructions executable by a runtime system to cause the runtime system to enforce the synchronization strategies of the parallel regions, the runtime system uses round-robin and sequence numbers and pulses as the synchronization strategy as instructed by the compiler, the runtime system performs a shuffle when two adjacent parallel regions include incompatible keys.