Scheduling multiple tasks in distributed computing system to avoid result writing conflicts

A system for processing a plurality of tasks, the system comprising:

• a first data processor configured to
• process a first set of tasks of the plurality of tasks, and
• write a result of the processing of the first set of tasks, to a system memory, in units of a first number of cache lines at a time;
• a second data processor configured to
• process a second set of tasks of the plurality of tasks, and
• write a result of the processing of the second set of tasks, to the system memory, in units of a second number of cache lines, different than the first number of cache lines;
• and
• a scheduler configured to avoid a potential result writing conflict due to the first number being different than the second number, by
• provisionally assigning the first set of tasks to the first data processor and the second set of tasks to the second data processor,
• analyzing how the second data processor is expected to write a result of the second set of tasks to the system memory to identify a potential result writing conflict, and
• in response to identifying the potential result writing conflict, adjusting the provisional assigning of the second set of tasks to the second data processor to prevent the potential results writing conflict;
• wherein the preventing of the potential result writing conflict prevents the result of processing of the second set of tasks from incompletely filling the second number of cache lines which would result in an unfilled portion of the second number of cache lines containing indeterminate data that would overwrite results from processing the first set of tasks.

A system for processing a plurality of tasks, the system comprising:

• a first processor configured to process tasks, store results of the processing in a first local buffer, and write contents of the first local buffer to a memory in units of blocks of a first block size;
• a second processor configured to process tasks, store results of the processing in a second local buffer, and write contents of the second local buffer to the memory in units of blocks of a second block size that is larger than the first block size;
• and
• a scheduler configured to, before the plurality of tasks is processed by the first processor and the second processor,
• analyze a provisional distribution of the plurality of tasks between the first processor and the second processor, in which a first set of the plurality of tasks is provisionally assigned to the first processor for processing to yield first results and a second set of the tasks is provisionally assigned to the second processor for processing to yield second results,
• determine that the provisional distribution will result in a block, that is written from the second local buffer to the memory, being only partially filled with the second results, and
• in response to the determination, adjust the provisional distribution of the plurality of tasks between the first processor and the second processor to avoid a potential result writing conflict due to the first block size being different than the second block size.