Cloud Workload Placement via Server-Storage Calibration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current cloud computing solutions often lead to suboptimal configurations by making compute and storage placement decisions independently, resulting in inadequate bandwidth between servers and storage nodes.
Innovation Solution
An empirical approach is introduced to calibrate cloud computing environments by determining different combinations of servers and storage units, deploying a virtual machine to generate a workload, and taking performance measurements to assess connection quality, which are then used to optimize future workload placement.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If compute and storage placement decisions are made independently, then the placement process is simple and fast, but the configuration becomes suboptimal with inadequate bandwidth between servers and storage nodes
Solution Approach 1:
The patent combines compute placement and storage placement decisions into a unified joint optimization process. The system evaluates server-storage unit combinations holistically, considering bandwidth constraints and performance requirements together rather than separately, thereby achieving optimal configurations that satisfy both computational and storage demands simultaneously.
Solution Approach 2:
The system performs preliminary characterization of the cloud environment by measuring bandwidth between servers and storage units before making placement decisions. This advance knowledge of connection qualities enables the joint optimization algorithm to make informed placement decisions that ensure adequate bandwidth, avoiding suboptimal configurations from the outset.
2Reliability
If joint optimization of compute and storage placement is performed, then configuration quality improves with adequate bandwidth, but the complexity of the placement process increases
Solution Approach 1:
The patent replaces complex manual or rule-based joint optimization processes with a machine learning model. The trained model directly predicts optimal server-storage unit combinations based on workload characteristics and environment parameters, eliminating the need for complex real-time optimization algorithms while maintaining high configuration quality.
Solution Approach 2:
The system creates a virtual representation or model of the cloud environment including server capacities, storage unit characteristics, and bandwidth measurements. This digital twin or simulated environment allows the joint optimization algorithm to evaluate placement options and train machine learning models without affecting the actual production system, reducing operational complexity.
3Measurement precision
If performance measurements are taken for all different combinations of servers and storage units, then accurate calibration is achieved, but the time and resources required for measurement increase
Solution Approach 1:
The patent employs sampling techniques where performance measurements are taken for a representative subset of server-storage unit combinations rather than exhaustively measuring all possible pairs. The system strategically selects combinations that provide sufficient information for accurate calibration, achieving acceptable precision with reduced measurement time and resource consumption.
Solution Approach 2:
The system performs preliminary characterization by measuring bandwidth and performance metrics for all server-storage unit combinations in advance during system setup or maintenance windows. This pre-collected data is stored and used for future placement decisions, eliminating the need for repeated real-time measurements while maintaining accurate calibration information.
Data Source
AI summary
In general, embodiments of present invention provide an approach for calibrating a cloud computing environment. Specifically, embodiments of the present invention provide an empirical approach for obtaining end-to-end performance characteristics for workloads in the cloud computing environment (hereinafter the “environment”). In a typical embodiment, different combinations of cloud server(s) and cloud storage unit(s) are determined. Then, a virtual machine is deployed to one or more of the servers within the cloud computing environment. The virtual machine is used to generate a desired workload on a set of servers within the environment. Thereafter, performance measurements for each of the different combinations under the desired workload will be taken. Among other things, the performance measurements indicate a connection quality between the set of servers and the set of storage units, and are used in calibrating the cloud computing environment to determine future workload placement. Along these lines, the performance measurements can be populated into a table or the like, and a dynamic map of a data center having the set of storage units can be generated.