Cloud Scheduler State Reduction for Temporarily Available Resources
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Cloud computing systems face challenges in managing stateful workloads on temporarily available resources, as they risk data loss when these resources become unavailable, limiting users to executing only stateless workloads to avoid data loss.
Innovation Solution
A scheduler in the cloud computing system receives state measurements from scheduled units and determines if they exceed the safe state capacity, prompting the units to perform state-reduction operations such as compression or upload to ensure data can be saved within a predefined time, allowing stateful workloads to be executed safely on temporarily available resources.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If stateful workloads are executed on temporarily available resources, then resource utilization is improved, but data loss risk increases
Solution Approach 1:
The system performs preliminary actions by monitoring the state size of workloads before temporarily available resources become unavailable. The scheduler detects when state size approaches a threshold and proactively triggers state-reduction operations (compression, upload, or eviction) before the resources are revoked, ensuring data is preserved in advance rather than reacting after loss occurs.
2Reliability
If state-reduction operations are performed continuously, then data safety is improved, but system complexity increases
Solution Approach 1:
The system applies different state-reduction operations to different workloads based on their specific characteristics. The scheduler evaluates each workload's state size and selects appropriate operations (compression for compressible data, upload for critical data, eviction for non-critical data) rather than applying a uniform approach to all workloads, thereby managing complexity through targeted local actions.
Solution Approach 2:
The scheduler automatically monitors state sizes and triggers state-reduction operations without requiring manual intervention or complex external control systems. The system self-manages the complexity by having the scheduler continuously assess workload states and autonomously decide when and how to reduce state, simplifying the overall system architecture while maintaining data safety.
3Reliability
If state measurements are monitored frequently, then data loss prevention is improved, but energy consumption increases
Solution Approach 1:
The scheduler performs state measurements periodically rather than continuously, checking state sizes at regular intervals or at trigger points such as before resource revocation. This periodic monitoring approach provides sufficient data to prevent data loss while avoiding the excessive energy consumption associated with continuous real-time monitoring of all workload states.
Data Source
AI summary
Stateful workloads executing on temporarily available resources of a cloud computing system can be managed. For example, a computing device can receive, from a scheduled unit executing on a temporarily available resource of a compute node in a cloud computing system, a state measurement indicating a size of a state of the scheduled unit. The computing device can receive, from the compute node, a safe state capacity that is a size of data capable of being saved in response to the temporarily available resource being unavailable. The computing device can determine the state measurement for the scheduled unit exceeds the safe state capacity. In response to determining the state measurement exceeds the safe state capacity, the computing device can transmit a command to the scheduled unit to perform a state-reduction operation.