HPC Load Distribution via Grouped Representative Nodes
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In high-performance computing (HPC) systems, the existing load distribution methods for parallel processing lead to increased standby times and overheads due to sequential access to shared task lists, especially as the number of nodes increases, resulting in inefficiencies during tasks like virtual screening.
Innovation Solution
A management device classifies nodes into groups, selects a representative node for each group, and has it generate a team task list from a shared file, allowing other nodes to acquire tasks from a team task list, thereby reducing standby times and improving task list generation efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If all nodes sequentially access the shared task list to acquire tasks, then task distribution is achieved, but standby time increases and system efficiency deteriorates as the number of nodes increases
Solution Approach 1:
The patent divides all nodes into multiple groups, with each group having a dedicated representative node. Each representative node independently accesses the shared task list for its group, segmenting the single access point into multiple concurrent access points. This reduces the standby time for task acquisition and improves overall system productivity by allowing parallel task list generation across groups.
2Ease of operation
If each node independently generates its task list from the shared file, then task autonomy is improved, but the overhead of repeated sequential access to the shared file increases
Solution Approach 1:
The patent segments the node population into groups with dedicated representative nodes. Each representative node generates task lists for its group members by accessing the shared task list once, rather than every node accessing it independently. This maintains task list generation autonomy while dramatically reducing the overhead of repeated sequential access operations.
Solution Approach 2:
The representative node creates copies of the task list for distribution to other nodes in its group. Instead of each node independently reading from the shared file, the representative node reads once and distributes copies to group members, reducing access overhead while maintaining each node's ability to autonomously process its assigned tasks.
3Productivity
If the number of nodes is increased to handle larger scale processing, then processing capacity is improved, but the standby time for task list access increases proportionally
Solution Approach 1:
The patent segments the growing node population into multiple groups, each with its own representative node. This allows the system to scale to larger numbers of nodes while maintaining constant standby time for task list access within each group, as each group independently manages its own task list generation without being blocked by other groups.
Data Source
AI summary
A device includes a processor configured to: classify arithmetic processing devices that executes tasks in parallel by distributing loads into arithmetic processing device groups; select a representative arithmetic processing device; notify the representative arithmetic processing device of identification information of other arithmetic processing devices of an arithmetic processing device group to which the representative arithmetic processing device belongs; instruct the representative arithmetic processing device to acquire information regarding tasks to be executed by the arithmetic processing devices of the arithmetic processing device group from a first task list, and to generate a second task list; notify each other arithmetic processing devices of identification information of the representative arithmetic processing device; and instruct each other arithmetic processing device to acquire information regarding tasks to be executed by the representative arithmetic processing device and each other arithmetic processing device from the second task list, and to generate a third task list.