Tenant Latency Signal Analysis for Cloud Service Issue Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing cloud service monitoring solutions rely on a single percentile for latency detection, which is sensitive to dataset size and fails to consider the distribution of percentile latency values, leading to ineffective scaling and increased human effort in selecting criteria for latency-related issue detection.
Innovation Solution
A system that analyzes latency signals on a tenant-by-tenant basis, using a defined set of percentiles with tenant-specific models to determine latency baselines, considering the distribution of percentile latency values, and generates a latency health score vector to make binary health determinations, aggregating these across geographic regions for notification of potential issues.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If a single percentile is used for latency detection, then the monitoring approach is simple to implement, but it is severely sensitive to dataset size and does not effectively scale to various cloud services
Solution Approach 1:
The patent implements a universal monitoring framework that automatically selects and applies appropriate percentile thresholds based on service-specific characteristics. The system provides a default set of percentiles (P50, P75, P90, P95, P99) that can be universally applied across different cloud services, while also allowing customization for specific service requirements. This multi-functionality enables the same monitoring infrastructure to serve diverse cloud services effectively without requiring manual configuration for each service.
Solution Approach 2:
The system dynamically adjusts percentile parameters based on service characteristics and dataset size. Instead of using a fixed single percentile threshold, the monitoring solution evaluates multiple percentile values and selects the appropriate ones automatically. The system can change the percentile parameters (P50, P75, P90, P95, P99) depending on the specific cloud service being monitored, the volume of data available, and the detected latency patterns, thereby adapting to different scenarios without manual intervention.
2Measurement precision
If manual percentile selection is performed to ensure effective monitoring, then detection accuracy can be optimized for specific services, but a considerable amount of human effort is consumed
Solution Approach 1:
The monitoring system performs self-service by automatically selecting and configuring the appropriate percentile thresholds without requiring manual human intervention. The system analyzes service characteristics, data patterns, and latency distributions autonomously to determine the optimal percentile values for monitoring. This self-configuration capability maintains high detection accuracy while eliminating the time-consuming manual percentile selection process, allowing the system to adapt to new services automatically.
Solution Approach 2:
The system performs preliminary analysis of latency data patterns and service characteristics during the onboarding phase to pre-determine the appropriate percentile thresholds. By conducting this analysis in advance, the system prepares the optimal monitoring configuration before actual monitoring begins, ensuring detection accuracy is optimized from the start without requiring ongoing manual adjustment. The preliminary action includes evaluating historical data patterns and establishing baseline percentile values specific to each service.
3Adaptability or versatility
If latency monitoring is performed across multiple tenants with varying request patterns, then comprehensive coverage is achieved, but latency noise and variability increase significantly
Solution Approach 1:
The monitoring system segments the multi-tenant environment by creating tenant-specific monitoring instances that analyze latency data independently for each tenant. This segmentation allows the system to maintain comprehensive coverage across all tenants while avoiding the noise and variability that would result from aggregating all tenant data into a single monitoring view. Each tenant's latency patterns are evaluated separately, enabling the system to account for individual request patterns and complexities without cross-tenant interference.
Solution Approach 2:
The system applies local quality by customizing the monitoring parameters and percentile thresholds specific to each tenant's characteristics and requirements. Instead of using a uniform monitoring approach across all tenants, the system adapts the monitoring strategy to match each tenant's specific latency patterns, request types, and performance expectations. This localized approach reduces noise by focusing on tenant-specific anomalies rather than being affected by variations in other tenants' workloads.
Data Source
AI summary
The techniques describe effective detection of latency-related issues for a cloud service operating in a distributed computing environment. To detect the latency-related issues, a system first determines baseline latency behavior at the tenant level (e.g., on a tenant-by-tenant basis) and compares a tenant's current latency behavior to the baseline latency behavior. If the comparison yields that the current latency behavior for the tenant is following the baseline latency behavior, the tenant is deemed healthy. However, if the comparison yields that the current latency behavior for the tenant is not closely following the baseline latency behavior, the tenant is deemed unhealthy. Once the system has made binary health determinations for various tenants on a tenant-by-tenant basis, the system is configured to aggregate the unhealthy determinations across a group of tenants to determine whether the cloud service is experiencing latency-related issues.