Auto-scaling Management Apparatus for Workload Overload Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing auto-scaling technologies in information processing systems often fail to accurately determine when to scale out resources, leading to wasteful resource allocation and prolonged response times due to incorrect identification of workload overload, especially in multi-tier systems where delays can be caused by factors other than workload overload.
Innovation Solution
A management apparatus that compiles concurrent process counts and processing times, calculates representative values based on subsets of processing times, and determines the need for resource addition by comparing newly compiled processing times with representative values, avoiding unnecessary resource allocation by distinguishing between workload overload and other causes of delay.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of time
If a scale-out action is performed when response time exceeds a predetermined threshold, then the number of information processing apparatuses is increased to reduce processing load, but resources are wasted when the delay is caused by factors other than workload overload
Solution Approach 1:
The patent replaces the simple threshold-based mechanical trigger with an AI-based diagnostic system that analyzes multiple parameters (CPU utilization, memory usage, network I/O, disk I/O, queue depth) to determine the root cause of delays. This substitution allows the system to distinguish between workload-related delays and system-related delays, preventing unnecessary scale-out actions when the cause is not workload overload.
Solution Approach 2:
The patent changes the decision parameter from a single response time threshold to a multi-parameter analysis including CPU utilization rate, memory usage rate, network I/O rate, disk I/O rate, and queue depth. By monitoring these additional parameters, the system can identify whether the response time delay is caused by workload overload or by system issues such as communication congestion or resource contention.
2Loss of time
If the number of information processing apparatuses is increased to reduce processing load, then response times are expected to decrease, but the scale-out action is ineffective when delays are caused by system failures or communication congestion
Solution Approach 1:
The patent replaces the ineffective mechanical scale-out trigger with an intelligent diagnostic system that identifies the root cause of delays before triggering scale-out. The system analyzes CPU utilization, memory usage, network I/O, disk I/O, and queue depth to determine whether the delay is workload-related, ensuring that scale-out actions are only performed when they will be effective.
Solution Approach 2:
The patent implements a feedback mechanism where the system continuously monitors multiple performance parameters and uses this information to adjust its decisions. By analyzing the relationship between these parameters and response times, the system learns to predict whether scale-out actions will be effective, improving the reliability of the auto-scaling mechanism.
3Productivity
If auto scaling is implemented based on CPU utilization or response time threshold exceedance, then the number of information processing apparatuses is adjusted, but the system cannot appropriately determine whether scale-out is effective when response time exceeds threshold
Solution Approach 1:
The patent expands the measurement parameters from simple CPU utilization or response time threshold to a comprehensive set including CPU utilization rate, memory usage rate, network I/O rate, disk I/O rate, and queue depth. This multi-parameter approach enables precise determination of whether response time exceedance is caused by workload overload or other factors, improving the accuracy of overload detection.
Solution Approach 2:
The patent replaces the simple threshold-exceedance detection mechanism with an AI-based analysis system that evaluates multiple parameters simultaneously. This substitution allows the system to accurately distinguish between genuine workload overload and other causes of response time delays, enabling more precise measurement and decision-making.
Data Source
AI summary
Each time an information processing apparatus has executed a requested process, a compiling unit compiles, in a memory unit, a concurrent process count indicating the number of processes being executed by the apparatus at the start of the process in association with a processing time of the process. For each concurrent process count, a calculating unit calculates a representative value based on a subset out of a set of processing times associated with the concurrent process count. The subset is composed of processing times more than a representative value calculated based on a subset of processing times associated with a concurrent process count less than the concurrent process count. A determining unit determines need for an addition to resources based on a newly compiled processing time and a representative value of a concurrent process count associated with the newly compiled processing time.