Have a lot of computing tasks, if a node to calculate, so need to compute over ten hours, so I need a queue management program,
A task management server A, the computing tasks are submitted to A task management server, A record the task in A queue,
And several other B class node, which is designed to handle computing tasks of computing nodes B1, B2, B3... Communication, communication once per minute, in order to query whether the node B free
If B1 idle, it will no longer be assigned to B1 task; If B2 idle, A computing tasks in the queue first task assigned to B2, so computing tasks in the queue the second task is to become the first; If this first B3 free the new task assigned to B3, then had the next computing tasks will be to first... And so on. If all of the class B nodes are busy, then suspend allocation, distribution of computing tasks in the task queue is not first,
You may have guessed, this is a small cluster of AI data analysis, the only topic type and quantity, and due to the nature of the title, in fact is not convenient to multiple parallel class B node, each node B alone will probably be better run, of course does not rule out future might involve multiple parallel class B node,
I need this function under Linux has a slurm is possible, but because I have to run the application is Windows only, so can't use the Linux + slurm, settle for second best for management under the Windows program
As to the storage of data, the other has A class C server, provides the Shared directory, and then let the A and B class server through the drive mapping to use C stored on the server,
I want to ask, have Windows comes with a similar software, or have any off-the-shelf software can realize the calculation task management function,
CodePudding user response:
Distributed computing, Windows should have professional software, never used