Dynamic Task Scheduling via Data Locality Tracking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In parallel computing systems, existing scheduling methods fail to efficiently distribute tasks across multiple processors due to non-uniform memory access times, leading to performance losses, especially when data is accessed repeatedly, as conventional heuristics lack data location information and require complex source code modifications.
Innovation Solution
A method that processes source code with parallel lambda functions to derive data location information, allowing tasks to be executed on processor cores associated with the memory units storing the required data, thereby optimizing memory access and reducing complexity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If conventional heuristics are used in the scheduler to distribute tasks, then tasks can be scheduled without complex source code modifications, but the scheduler lacks data location information leading to suboptimal memory access performance
Solution Approach 1:
The patent introduces an intermediary component (the scheduling system with data location tracking) that bridges the gap between task execution and memory location information. The system automatically tracks and maintains data location information in memory without requiring source code modifications, thereby resolving the contradiction between ease of scheduling and memory access performance.
2Productivity
If explicit data location information is added to source code to influence scheduling, then memory access performance can be optimized, but the source code complexity increases and becomes harder to maintain
Solution Approach 1:
The system implements self-service by automatically tracking data location information through memory access patterns without requiring any explicit annotations or modifications to the source code. The scheduling system autonomously gathers and utilizes data location information, thereby optimizing memory access performance while keeping source code simple and maintainable.
3Productivity
If tasks are mapped to remote processors, then load balancing can be improved, but memory access time increases significantly due to non-uniform memory access
Solution Approach 1:
The patent implements dynamic task scheduling that adapts to data location information in real-time. Instead of static load balancing assignments, the system dynamically adjusts task-to-processor mapping based on current data locations, thereby reducing memory access time while maintaining effective load distribution across processors.
4Adaptability or versatility
If data is loaded from remote memory, then tasks can execute on any processor, but system performance is significantly reduced
Solution Approach 1:
The system performs preliminary action by pre-loading or pre-positioning data in memory locations that are optimally accessible to the processors that will need them. By anticipating data access patterns and preparing data locations in advance, the system maintains processor flexibility while avoiding performance degradation from remote memory accesses.
Data Source
AI summary
A method for scheduling tasks to processor cores of a parallel computing system may include the steps of processing a source code which comprises at least one parallel lambda function having a function body called by a task and having a capture list specifying the data structures accessed in the function body of said parallel lambda function and used to derive data location information; executing the task calling said function body on the processor core which is associated to a memory unit of the parallel computing system where the data of the data structures specified by said capture list is stored, wherein the memory unit is selected or localized based on the derived data location information.