Storage Response Time Distribution Estimation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Predicting response times for storage systems is challenging due to the complexity of telemetry data, requiring high-quality data selection, accurate modeling, informative predictions, and fast configuration evaluation to ensure adequate performance under proposed workloads.
Innovation Solution
A method involving the aggregation of workload characteristics and storage system parameters to train a probabilistic mixture model, allowing for the estimation of response time distributions using a deep neural network, enabling accurate and efficient prediction of storage system performance for proposed workloads.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If operational testing is performed to evaluate storage system performance, then prediction accuracy is improved, but time consumption and testing complexity increase
Solution Approach 1:
The system performs preliminary actions by collecting and storing telemetry data from multiple storage systems during normal operation before actual performance evaluation is needed. This pre-collected data is then used to train machine learning models, enabling accurate predictions without requiring time-consuming operational testing at the time of evaluation.
Solution Approach 2:
The system creates copies of operational characteristics by collecting telemetry data from multiple different storage systems and workloads. These copied data patterns are used to train models that can predict performance of proposed storage systems without actually running operational tests on them.
2Measurement precision
If detailed telemetry data is collected from multiple storage systems, then model training accuracy is improved, but data processing complexity increases
Solution Approach 1:
The system extracts only the essential and relevant features from the collected telemetry data, such as response times, throughput, and key operational parameters. By taking out only the necessary information rather than processing all raw data, the system maintains high model training accuracy while reducing data processing complexity.
Solution Approach 2:
The system merges data from multiple different storage systems and workload types into a unified training dataset. By combining these diverse data sources and aggregating their operational characteristics, the system improves model generalization and accuracy while managing complexity through standardized data processing pipelines.
3Adaptability or versatility
If machine learning models are trained on diverse workload characteristics, then prediction versatility is improved, but computational resources required increase
Solution Approach 1:
The system performs preliminary training of machine learning models using diverse workload characteristics from multiple storage systems. This pre-training enables the models to be versatile and adaptable to different workloads without requiring intensive computational resources at the time of actual performance prediction.
Solution Approach 2:
The system uses parameter changes by training models on varied workload parameters and characteristics during the training phase. This allows the models to learn diverse patterns and become versatile in predicting performance across different workload types, while the actual prediction process requires minimal computational resources since the learning is already complete.
Data Source
AI summary
A distribution of response times of a storage system can be estimated for a proposed workload using a trained learning process. Collections of information about operational characteristics of multiple storage systems are obtained, in which each collection includes parameters describing the configuration of the storage system that was used to create the collection, workload characteristics describing features of the workload that the storage system processed, and storage system response times. For each collection, workload characteristics are aggregated, and the storage system response information is used to train a probabilistic mixture model. The aggregated workload information, storage system characteristics, and probabilistic mixture model parameters of the collections form training examples that are used to train the learning process. Once trained, the learning process is used to provide a distribution of response times that would be expected from a storage system having a proposed configuration when processing a proposed workload.