Parallel Computing Compression Mapping for Inter-Core Data Transfer
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Large-scale parallel computing operations lead to a significant increase in communication data between processors, resulting in performance drops due to increased data transfer times, which existing technologies fail to efficiently address by determining compression schemes in real-time.
Innovation Solution
An electronic device predetermines compression schemes based on dominant data patterns analyzed before execution, applying block floating point-based and zero-value-aware-based compression schemes during communication phases to reduce data transfer between cores.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Speed
If the number of processors is increased to execute large-scale operations in parallel, then the execution speed of the application is improved, but the amount of communication data between processors increases significantly, causing performance drop
Solution Approach 1:
The patent applies preliminary action by analyzing application code before execution to identify communication phases and predetermining compression schemes for each phase. This pre-analysis creates a mapping between communication phases and optimal compression algorithms, which is then applied during runtime without requiring real-time decision-making. The compression schemes are prepared in advance based on static analysis of the application's communication patterns.
Solution Approach 2:
The patent changes the parameter of data representation by applying different compression schemes to different communication phases. Instead of using a single compression method or no compression, the system dynamically selects from multiple compression schemes (such as lossless compression, lossy compression, or no compression) based on the characteristics of each communication phase, thereby optimizing the balance between communication volume and data accuracy.
2Productivity
If real-time compression scheme determination is implemented, then data transfer efficiency is improved, but the complexity of the system increases and real-time decision-making becomes challenging
Solution Approach 1:
The patent resolves the complexity issue by performing compression scheme selection in advance during a static analysis phase. The system analyzes the application code before execution to identify all communication phases and determines the optimal compression scheme for each phase beforehand. This eliminates the need for complex real-time decision-making logic during runtime, as the compression scheme for each communication phase is already predetermined and stored in a lookup table.
Solution Approach 2:
The patent segments the application execution into distinct communication phases and operation phases. By identifying and separating communication phases (where data is exchanged between processors) from operation phases (where computations are performed), the system can apply compression selectively only during communication phases. This segmentation simplifies the overall system design by localizing the compression functionality to specific phases rather than requiring continuous compression monitoring throughout the entire execution.
3Quantity of substance
If compression schemes are applied to all communication phases, then the amount of data transferred is reduced, but the overhead of compression and decompression may affect execution performance
Solution Approach 1:
The patent optimizes the trade-off between compression benefits and overhead by selecting from multiple compression schemes with different characteristics. For communication phases where data patterns are highly compressible, aggressive compression schemes are applied. For phases where data is already compact or where compression overhead would be excessive relative to the benefit, the system may select lighter compression or no compression at all. This parameter-based selection allows flexible optimization of the compression overhead versus data reduction ratio.
Solution Approach 2:
The patent applies compression selectively rather than universally. By analyzing the characteristics of each communication phase, the system applies compression only when it provides net benefit. Some communication phases may receive full compression treatment, while others may receive minimal or no compression if the overhead would exceed the benefits. This partial application of compression optimizes the overall performance by avoiding unnecessary compression operations.
Data Source
AI summary
Disclosed are electronic devices with predetermined compression schemes for parallel computing and methods thereof. An example electronic device includes cores of one or more processors, one or more memories storing instructions configured to, when executed by the cores, configure the cores to perform operations of an application executed on the electronic device, the operations including communication phases that communicate data between the cores, wherein the application includes, prior to execution of the application on the electronic device, predetermined information associating the communication phases with respective compression schemes, and apply the compression schemes corresponding to the communication phases according to the predetermined information to compress the data of the communication phases that is exchanged between the cores when executing the application.