Adaptive Compression Entropy Threshold Accelerator
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional systems degrade system throughput and latency by attempting to compress data that is not compressible, leading to output data larger than the input data, which is then unnecessarily processed and transmitted back to the accelerator device.
Innovation Solution
An accelerator device computes the entropy value of input data and compares it to a threshold; if the entropy value exceeds the threshold, it refrains from compressing the data, instead generating stored blocks or returning an indication that the data will not be compressed, thereby avoiding unnecessary processing and transmission.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the accelerator device compresses data without checking compressibility, then compression processing is performed, but the output data size increases and system throughput degrades
Solution Approach 1:
The patent applies preliminary action by computing the entropy value of the input data before performing compression. The entropy computation is performed in advance to determine whether the data is suitable for compression, preventing unnecessary compression operations that would increase output size and degrade throughput.
Solution Approach 2:
The patent uses feedback by comparing the computed entropy value against a threshold to make a decision about whether to compress. This feedback mechanism allows the system to adapt its behavior based on the characteristics of the input data, compressing only when entropy indicates potential size reduction.
2Productivity
If the accelerator device compresses uncompressible data, then processing is performed, but latency increases due to unnecessary processing and memory transmission
Solution Approach 1:
The entropy computation is performed as a preliminary action before compression. This early assessment prevents the system from entering unnecessary compression and memory transmission workflows, thereby reducing latency and processing time for uncompressible data.
Solution Approach 2:
The patent extracts the decision-making step (entropy computation and threshold comparison) from the compression workflow. By separating the compressibility assessment from the actual compression operation, the system can bypass unnecessary processing steps for uncompressible data, reducing overall processing time.
3Productivity
If the accelerator device attempts to compress all data, then compression operations are executed, but compression ratio deteriorates when output is larger than input
Solution Approach 1:
The patent employs feedback by using the entropy value to guide the compression decision. The entropy computation provides information about the data's compressibility, and this feedback determines whether compression should proceed, ensuring that compression is only applied when it will achieve a favorable compression ratio.
Solution Approach 2:
The patent changes the parameter being monitored from simple data size to entropy value. By using entropy as the decision parameter, the system can accurately identify compressible data versus uncompressible data, thereby maintaining optimal compression ratios and avoiding cases where compressed output exceeds input size.
Data Source
AI summary
An accelerator device may access an input data chunk to be compressed by the accelerator device. The accelerator device may access an entropy value for the input data chunk. The accelerator device may compress the input data chunk or return an indication that the input data chunk will not be compressed based on the entropy value and an entropy threshold.