Improving efficiency of a global barrier operation in a parallel computer

摘要

Performing a global barrier operation in a parallel computer that includes compute nodes coupled for data communications, where each compute node executes tasks, with one task on each compute node designated as a master task, including: for each task on each compute node until all master tasks have joined a global barrier: determining whether the task is a master task; if the task is not a master task, joining a single local barrier; if the task is a master task, joining the global barrier and the single local barrier only after all other tasks on the compute node have joined the single local barrier.

主权项

1. A method of performing a global barrier operation in a parallel computer, the parallel computer comprising a plurality of compute nodes, the compute nodes coupled for data communications, each compute node executing a plurality of tasks, with one task on each compute node designated as a master task, the method comprising:
for each task on each compute node until all master tasks have joined a global barrier: determining whether the task is a master task, wherein each task includes an indicator indicating whether the task is or is not a master task; if the task is not a master task, joining a single local barrier on a compute node of the plurality of compute nodes; if the task is a master task, joining both the single local barrier on the compute node and the global barrier only after all other tasks on the compute node have joined the single local barrier on the compute node; and wherein joining the single local barrier includes atomically incrementing a value of a counter, which tracks tasks that joined the single local barrier, and a number of times equivalent to a result of a difference between a total number of tasks joining the single local barrier and a replacement value, the replacement value comprising a power-of-two greater than or equal to the total number of tasks joining the single local barrier on the compute node.