Event Digest Scheduling for Scalable Distributed Host Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Hyperscale data centers face scalability challenges due to exponential increases in resource and storage requirements for telemetry data, leading to inefficiencies in monitoring, scheduling, and error handling, particularly with conventional in-band approaches that introduce overhead and latency.
Innovation Solution
The implementation of an event digest process that generates a compact representation of platform metrics, allowing for out-of-band monitoring and management, which reduces the need for OS or VMM involvement, thereby improving scalability and performance by aggregating telemetry points into abstracted states and enabling efficient scheduling and error detection.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional in-band monitoring approaches are used to collect telemetry data from each monitored entity, then comprehensive monitoring coverage is achieved, but the amount of data transferred to the controller increases exponentially with the number of monitored entities
Solution Approach 1:
The patent segments the telemetry data collection process by introducing intermediate aggregation points (edge devices, regional controllers) that collect and summarize data from multiple monitored entities before transmitting to the central controller. This hierarchical segmentation reduces the total data transfer volume while maintaining comprehensive monitoring coverage through distributed data collection points.
Solution Approach 2:
The patent extracts only the essential monitoring information from complete telemetry data sets by implementing event-based filtering and anomaly detection at the edge devices. Only events exceeding thresholds or representing significant changes are extracted and transmitted to the controller, rather than transmitting all raw telemetry data, thereby reducing data transfer volume while preserving monitoring effectiveness.
2Speed
If high polling rates are implemented to achieve fast responsiveness in telemetry processes, then monitoring responsiveness is improved, but the processing cycles and computational resources required increase exponentially
Solution Approach 1:
The patent implements periodic polling with adaptive intervals rather than continuous high-rate sampling. Polling rates are dynamically adjusted based on system state - higher rates during periods of change or anomaly, and lower rates during stable conditions. This periodic action with variable frequency maintains monitoring responsiveness while significantly reducing average processing cycle consumption.
Solution Approach 2:
The patent enables monitored entities and edge devices to perform self-monitoring and local anomaly detection, generating events only when predefined thresholds are exceeded. This self-service approach eliminates the need for continuous centralized polling, allowing the system to maintain fast responsiveness to actual events while minimizing routine processing cycle consumption through event-driven rather than time-driven architecture.
3Productivity
If the number of monitored entities in a data center is increased to support hyperscale operations, then system capacity is improved, but the storage requirements for telemetry data increase exponentially
Solution Approach 1:
The patent merges telemetry data from multiple monitored entities at intermediate aggregation points before storing in the central system. By combining and deduplicating data at the edge, the system supports increased numbers of monitored entities (improving productivity) while reducing the volume of unique data requiring centralized storage. Aggregated summaries and consolidated event logs reduce storage requirements compared to storing complete individual telemetry streams from each entity.
Solution Approach 2:
The patent implements selective data retention policies where routine, normal telemetry data is discarded or stored locally at edge devices, while only significant events, anomalies, and aggregated summaries are retained in centralized storage. This discarding of redundant data allows the system to scale to more monitored entities without proportionally increasing centralized storage requirements, as the majority of telemetry data from additional entities follows similar patterns and can be locally managed.
4Loss of information
If detailed telemetry data from all monitored entities is collected and transmitted to the controller, then comprehensive system visibility is achieved, but the resource consumption and overhead increase exponentially
Solution Approach 1:
The patent applies different data collection and transmission qualities to different parts of the system based on their importance and characteristics. Critical monitored entities and key performance indicators receive higher monitoring fidelity and more frequent reporting, while less critical entities use lower-fidelity monitoring. This local quality differentiation maintains comprehensive system visibility for important parameters while reducing overall resource consumption by adapting monitoring intensity to local requirements.
Data Source
AI summary
Techniques and apparatus for managing a distributed computing environment using event digests are described. In one embodiment, for example, an apparatus may include at least one memory, and logic for a system manager, at least a portion of the logic comprised in hardware coupled to the at least one memory, the logic to determine a workload to schedule, access an event digest associated with a plurality of compute hosts, the event digest comprising event digest values determined using out-of-band information, determine metrics from the event digest, generate at least one host weight for at least a portion of the plurality of compute hosts based on the metrics, identify at least one candidate host from the portion of the plurality of compute hosts based on the at least one host weight, and schedule the workload on the at least one candidate host. Other embodiments are described and claimed.