Pooled Input/Output Dispatch Layer for Write-Heavy Traffic
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Distributed computing systems face limitations in flexibility and scalability due to resource coupling, and existing I/O request dispatching methods often lead to inefficiencies, particularly with write-intensive flows affecting waiting times and overall performance.
Innovation Solution
Decoupling resources into pooled layers of same-type resources and using a dispatch layer that assigns operations based on weighted moving averages and traffic levels, prioritizing read-intensive flows, and employing constrained or unconstrained random assignment to manage traffic spikes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If resources are coupled together in traditional computing systems, then resource allocation is simplified, but flexibility and scalability are limited
Solution Approach 1:
The patent segments resources into separate pooled layers (compute resources, storage resources, network resources) that can be independently managed and allocated. This segmentation allows the system to scale individual resource types independently, improving flexibility without requiring complete system redesign.
Solution Approach 2:
The dispatch layer acts as an intermediary between resource pools and computing tasks, intelligently matching requirements with available resources. This mediator enables flexible resource allocation and scalability without direct coupling between resource types.
2Loss of time
If I/O requests are queued based on fairness principles, then all requests are treated equally, but write-intensive flows increase waiting times
Solution Approach 1:
The patent applies different quality of service characteristics to different I/O flows based on their nature. Read-intensive flows receive priority handling with shorter waiting times, while write-intensive flows are managed separately. This local differentiation optimizes overall system performance by addressing the specific needs of each flow type.
Solution Approach 2:
The system dynamically changes dispatching parameters based on traffic conditions, using weighted moving averages to adapt priority levels and allocation strategies. This allows the system to respond to changing workload characteristics and minimize waiting times for critical operations.
3Measurement precision
If dispatching decisions use comprehensive historical performance data, then decision accuracy improves, but computing overhead increases
Solution Approach 1:
The patent implements periodic sampling of performance metrics using weighted moving averages within defined time windows. This periodic measurement approach captures relevant performance trends while avoiding the computational burden of continuously processing all historical data, balancing accuracy with efficiency.
4Productivity
If I/O flows are assigned to specific resources deterministically, then resource utilization is optimized, but traffic spikes can overload individual resources
Solution Approach 1:
The patent implements dynamic dispatching where the assignment of I/O flows to resources changes based on real-time conditions. The system monitors resource status and traffic patterns, adjusting assignments to balance utilization with overload prevention. This dynamic approach allows the system to adapt to traffic spikes by redistributing loads across available resources.
Data Source
AI summary
Decoupled computing systems include layers of same-type computing resources, and include a dispatch layer to assign tasks from one layer to another, such as input and output (I/O) flows. The I/O flows can be assigned to particular computing resources of a layer based on a weighted moving average of performance data for the layer. When traffic is high, the assignment can include random assignment to some or all of the computing resources in the layer. The I/O flows can be split between read-intensive and write-intensive flows, with more read-intensive flows being assigned based on a pick ratio.