Distributed Workload Scheduling via Two-Phase Timer Orchestration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Scheduling workloads in distributed computing environments at scale and maintaining specified rates is challenging, as central computers can become overloaded, and distributing workloads across multiple instances leads to incorrect rates due to lack of synchronization.
Innovation Solution
A distributed two-phase algorithm utilizing an external timer service and local high-resolution timers to orchestrate time-critical workloads, breaking down workloads into batches and using local timers to synchronize execution, thereby avoiding network latency issues and allowing for scalable and stateless workload scheduling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If workload is scheduled on a central computer, then rate control is maintained, but the central computer becomes overloaded when scaling to very large workloads
Solution Approach 1:
The patent divides the centralized scheduling function into distributed scheduling units across multiple computing instances. Each instance independently schedules portions of the workload while maintaining coordination through standardized interfaces, eliminating the single point of failure and overload bottleneck inherent in centralized scheduling.
Solution Approach 2:
The patent introduces a workload scheduling service as an intermediary layer between the distributed computing instances and the workload management system. This service coordinates scheduling decisions across instances, maintains rate control policies, and enables scalable workload distribution without requiring a centralized scheduler.
2Productivity
If workload is distributed over multiple instances, then overload is avoided, but incorrect rates occur because each instance is not aware of what other instances are doing
Solution Approach 1:
The patent implements feedback mechanisms where distributed scheduling units exchange information about their scheduling decisions and current workload states. This feedback loop enables each instance to adjust its scheduling behavior based on the actions of other instances, maintaining coordinated rate control across the distributed system.
Solution Approach 2:
The patent creates a universal workload scheduling service that can be instantiated across multiple computing instances, each performing the same scheduling functions. This standardized multi-functional approach ensures consistent rate control behavior across all instances while maintaining distributed autonomy.
3Measurement precision
If centralized scheduling is used, then rate control is precise, but the system cannot scale to very large workload amounts
Solution Approach 1:
The patent transitions from a single-dimensional centralized scheduling model to a multi-dimensional distributed scheduling architecture. By adding the dimension of distribution across multiple instances while maintaining coordination through standardized services, the system achieves both precision and scalability simultaneously.
Data Source
AI summary
Systems and methods for managing rate-controlled and scheduled workloads in distributed computing systems. One example method provides a distributed two-phase algorithm, which utilizes an external timer service and local high-resolution timers to orchestrate time critical workloads on distributed computing systems. The first phase is a planning phase implemented as a recurrent process by means of repeated registrations to the external timer service. During the planning phase, a portion or all of the workload to be performed is broken down into small batches to be assigned to computing instances for execution at a later time. The second phase is a batch execution phase. Each batch planned during the planning phase includes fine-grained instructions as to when each item within the batch is to be executed. The computing instance receiving the batch uses local high-resolution timers to synchronize the execution of items relative to one another.