Cohesive Clustering for Virtualized Storage Latency
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In virtualized computing environments, existing host clustering methods like k-means clustering are resource-intensive and do not account for varying network latency, leading to unacceptable delays when virtual disks are mapped to hosts with high network latency, disrupting storage services.
Innovation Solution
The formation of cohesive clusters based on network latency, where hosts with low latency are grouped together to ensure that compute resources and storage resources are mapped within the same cluster, reducing latency and improving storage service efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If k-means clustering is used to partition hosts into predetermined clusters, then hosts can be organized into storage clusters, but significant computational resources are consumed due to the iterative nature of the algorithm
Solution Approach 1:
The patent changes the fundamental parameter of clustering from iterative distance-based assignment (k-means) to a direct graph-theory-based partitioning method. By representing hosts as nodes and network latencies as edge weights, the system performs non-iterative spectral clustering that avoids the computational overhead of repeated distance calculations and convergence checks inherent in k-means algorithms.
Solution Approach 2:
The patent replaces the mechanical iterative process of k-means clustering with a mathematical graph partitioning approach. Instead of repeatedly calculating distances and reassigning points, the system uses graph Laplacian eigen decomposition to directly determine optimal cluster assignments, substituting an iterative mechanical process with a direct mathematical solution.
2Ease of operation
If k-means clustering treats all hosts equally without considering physical disk locations, then clustering is simpler to implement, but network latency varies significantly causing unacceptable delays in storage service
Solution Approach 1:
The patent applies local quality by making cluster formation dependent on local network latency characteristics between specific host pairs. Instead of treating all hosts uniformly, the graph-based approach captures local connectivity patterns and latency variations, ensuring that hosts with low latency to each other are grouped together while maintaining implementation through standard graph algorithms.
Solution Approach 2:
The patent introduces dynamics by making the clustering configuration adaptive to actual network conditions. The graph representation allows the system to dynamically reflect changing network latencies, and cluster assignments can be recalculated based on current latency measurements, enabling the system to adapt to varying network conditions rather than relying on static predefined clusters.
Data Source
AI summary
An example method to provide a storage service in a virtualized computing environment may include obtaining network latency information associated with various hosts in the virtualized computing environment. In response to identifying a first network latency value that exceeds a threshold, wherein the first network latency is between the first host and a second host in the virtualized computing environment, the method includes placing the second host in a first cohesive cluster consisting of the second host. In response to identifying a second network latency value being less than or equal to the threshold, wherein the second network latency value is between the first host and a third host in the virtualized computing environment, the method includes grouping the first host and the third host in a second cohesive cluster. The method includes initiating configuration of resources in the first cohesive cluster and the second cohesive cluster to support the storage service.