Workload Dispatcher Share Calculation for Sysplex Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Inefficient workload balancing in a parallel sysplex can lead to system overload, as existing methods fail to accurately distribute workload items across multiple systems, particularly in cases of capacity shortages or outages, potentially causing infinite queue growth and resource congestion.
Innovation Solution
A method for calculating a dispatcher's share of workload items in a workload manager, involving the use of an arbitrator that distributes workload items based on CPU consumption measurements, capacity values, and queue lengths across multiple systems, ensuring optimal distribution and preventing overload by normalizing shares and using arithmetic operations to combine relative queue lengths.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If workload is distributed across multiple systems in a parallel sysplex, then system reliability and scalability are improved, but workload balancing becomes more complex and may lead to overload if not calculated correctly
Solution Approach 1:
The patent changes the parameters used for workload distribution from simple static metrics to dynamic calculations incorporating service units (CPU consumption measurements), capacity values, and queue lengths. The dispatcher's share is calculated using a formula that combines these parameters: D = min(R1, R2, R3) × (1 / (1 + max(V1, V2, V3))), where R represents capacity ratios and V represents relative queue lengths. This parameter transformation enables accurate real-time workload balancing across the sysplex cluster.
2Productivity
If workload distribution is based on simple metrics, then calculation is fast and simple, but accuracy is insufficient leading to queue growth and resource congestion
Solution Approach 1:
The patent implements feedback mechanisms by continuously monitoring service units consumed by workload items, capacity values of execution units, and queue lengths at dispatchers. These metrics are fed back into the dispatcher's share calculation formula, which adjusts distribution proportions dynamically. The arbitrator receives this calculated share information and uses it to make real-time routing decisions, creating a closed-loop feedback system that ensures accurate workload distribution while maintaining calculation efficiency through a standardized formulaic approach.
3Reliability
If capacity shortages or outages occur, then system resilience is tested, but existing methods fail to prevent infinite queue growth and congestion
Solution Approach 1:
The patent applies preliminary anti-action by proactively calculating and adjusting dispatcher's shares before capacity shortages lead to system failure. The formula D = min(R1, R2, R3) × (1 / (1 + max(V1, V2, V3))) inherently prevents overload by considering the minimum capacity ratio across all workload types and penalizing systems with high queue lengths through the max(V) term. This preemptive adjustment of distribution proportions prevents queue growth and congestion before they occur, rather than reacting after problems arise.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
The present invention provides a method of calculating a dispatcher s share (D) in a workload manager, the workload manager coupled to an arbitrator, the arbitrator coupled to a plurality of systems, each system comprising a dispatcher, each dispatcher coupled to a plurality of execution units, the arbitrator being adapted for receiving and distributing a flow of workload items to the dispatchers, the execution units being adapted for executing the workload items, the workload items being of at least three types, the method comprising the steps of: reading a total of service units (W) for each workload type across the plurality of systems, the service units being a value for measuring CPU consumption; reading a capacity (cap) value for each workload type on each system (SYS), the capacity value being indicative of a maximum of service units that a system can execute. The method further comprises: calculating a dispatcher s relative share (R) by dividing the capacity by the total of service units; obtaining a minimum value of the dispatcher s relative share of each system; calculating a relative queue length (V) for each system by dividing a queue length (q) of the workload items for each workload type on each system by the capacity of each workload type on each system; calculating the dispatcher s share (D) for each system by multiplying the minimum value and a first function of the relative queue length.