Parallel python is a python distributed computing module. Its function is to distribute tasks to the CPUs of each machine in the cluster in the form of functions and data. At the same time, parallel Python cluster construction supports auto discovery, which can easily build a self-organized distributed computing platform in the internal network. Parallel Python does not have DFS and can be understood as mapreduce.
It has been a while before learning pp. After practice, we can draw the following conclusions:
- It is not suitable for Distributed scenarios where a large amount of data is exchanged. It directly shows that the socket connection is automatically passed out and the remote machine is automatically disconnected from the server.
- Not SuitableProgram(The current version 1.6.1 has not been resolved)
- It is not suitable for the distribution of a large amount of computing, mainly because the server computing time is too long and will automatically disconnect from the client (control end ).
- Its controllability is not strong, and it cannot control the assignment of tasks. From its source code, we can see that tasks are allocated to the local idle CPU preferentially, then pop is used to process the idle CPU of any machine from the task queue.
Applicable scenarios (personal understanding ):
- Task Distribution, without the need for a large amount of data transmission; high computing programs and the computing time of each task is not long, the default is 20 s
- The distribution executable body does not have too complex dependencies between modules, functions, and variables.