Object Reassignment for Cluster Load Balancing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In multi-computing environments, existing technologies face challenges in efficiently balancing the assignment of objects across processing units, leading to uneven distribution of storage and load capacities, which can result in inefficiencies and potential capacity limitations.
Innovation Solution
A method is introduced that involves sorting objects by size or load to create an ordered sequence, then iteratively reassigning them between processing units to optimize storage and load capacity utilization, using threshold levels to determine suitable units and calculating statistical measures for improved distribution smoothness.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If objects are assigned to processing units using existing load balancing algorithms, then storage and load capacities are distributed across processing units, but the distribution becomes uneven leading to capacity limitations and inefficiencies
Solution Approach 1:
The patent applies preliminary action by sorting objects in descending order of size before assignment. This pre-ordering ensures that larger objects are assigned first to processing units with sufficient capacity, preventing later assignment failures and reducing the need for swap operations. The sorting criterion is calculated in advance based on object characteristics.
Solution Approach 2:
The patent changes the assignment parameter from simple round-robin or random distribution to size-based ordered assignment. By using object size as the primary sorting criterion and capacity thresholds as assignment boundaries, the system achieves smoother distribution that respects both storage capacity and load capacity constraints of processing units.
2Reliability
If processing units are assigned objects without considering capacity thresholds, then assignment is simple and fast, but processing units may exceed their storage or load capacity leading to system failures
Solution Approach 1:
The patent applies local quality by defining specific capacity thresholds for each processing unit based on its individual storage capacity and load capacity. Each processing unit has tailored thresholds that reflect its local characteristics, allowing the system to maintain reliability while handling heterogeneous processing units with different capabilities.
Solution Approach 2:
The system performs preliminary calculation of capacity thresholds for each processing unit before assignment begins. These thresholds are stored and reused during the assignment process, avoiding repeated complex calculations while ensuring capacity constraints are always respected. The thresholds act as pre-computed guardrails.
3Ease of operation
If objects are reassigned frequently to balance load, then distribution smoothness improves, but the number of swap operations increases reducing system efficiency
Solution Approach 1:
The patent performs preliminary sorting of objects by size in descending order before the assignment process. This pre-ordering ensures that larger objects are placed first on appropriate processing units, creating a more balanced distribution from the start and minimizing the need for subsequent swap operations to achieve balance.
Solution Approach 2:
The system changes the assignment strategy from capacity-agnostic methods to size-aware ordered assignment. By considering object size as a primary parameter and using it to determine assignment order and target selection, the system achieves smoother distribution with fewer swaps, as objects are placed in their final positions more accurately on first assignment.
Data Source
AI summary
Objects are assigned to processing units of a cluster of processing units, each of the objects having an object size and an object load, each of the processing units having a storage capacity and a load capacity. The method uses an initial object distribution as a starting point. The smoothness of the initial object distribution is improved by processing the objects in sequential order.