Application Cluster Capacity Calculation via Simulated Workload Testing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for determining the capacity of an application cluster, such as human estimation and single-machine workload pressure testing, are prone to inaccuracies due to external factors and deviations in data, especially when transaction traffic models differ significantly from normal states.
Innovation Solution
A method involving the setup of a simulated application cluster with multiple testing machines to perform overall workload pressure testing, adjusting the workload pressure value to reach a rate limit threshold, and calculating the application cluster capacity based on the average single-machine capability, which is dynamically adjusted using auto-scaling mechanisms and planned data-handling capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If single-machine workload pressure testing is used to determine application cluster capacity, then the testing process is simple and quick, but the accuracy of capacity calculation is poor due to external factors and idle VM resources causing overestimated machine capability
Solution Approach 1:
The patent segments the capacity testing process into multiple phases: initial capacity estimation using single-machine testing, followed by iterative refinement using the formula C_new = C_old × (1 - α) where α represents the adjustment factor based on observed deviations. This segmentation allows the system to start with a simple approach and progressively improve accuracy without requiring a completely complex testing system from the beginning.
Solution Approach 2:
The patent creates a virtual copy of the application cluster environment for capacity testing, using virtual machines to simulate the actual production environment. This allows capacity calculations to be performed in a controlled virtual setting that mirrors real-world conditions, improving measurement precision without the complexity of testing on actual production systems.
2Reliability
If human estimation is used to calculate application cluster capacity, then the process is simple and requires minimal resources, but the reliability of capacity determination is low due to subjective factors and lack of empirical data
Solution Approach 1:
The patent performs preliminary capacity estimation using automated workload pressure testing before final capacity determination. This preliminary action provides an initial capacity value that is then refined through iterative calculations, reducing the need for time-consuming manual estimation while establishing a reliable baseline for further optimization.
Solution Approach 2:
The patent implements a feedback mechanism where capacity calculations are continuously refined based on observed system performance and workload characteristics. The formula C_new = C_old × (1 - α) incorporates feedback from previous capacity estimates and actual system behavior, progressively improving reliability while maintaining efficient automated processing that reduces time loss compared to manual methods.
3Measurement precision
If one-machine workload pressure test is performed, then the capability of a single machine can be obtained, but other VM resources of the host machine remain idle causing overestimated machine capability and large deviations in capacity data
Solution Approach 1:
The patent merges multiple virtual machines into a single testing cluster to perform workload pressure testing. By combining the resources of multiple VMs, the system achieves more accurate machine capability measurements that reflect actual production environment conditions, while improving resource utilization efficiency as all VM resources participate in the testing process rather than remaining idle.
4Adaptability or versatility
If the transaction traffic model for large-scale promotion is used, then the capacity planning should reflect peak demand, but this model is very different from normal state traffic model affecting the accuracy of capacity estimation
Solution Approach 1:
The patent implements dynamic capacity estimation that adapts to different traffic models by adjusting the workload pressure testing parameters based on the specific traffic characteristics being evaluated. The system can switch between normal state and peak demand traffic models, applying appropriate weighting factors to generate capacity estimates that are accurate for the specific operational context rather than being biased toward one traffic pattern.
Data Source
AI summary
Provided in the disclosed embodiments are a method and an apparatus for obtaining a capacity of an application cluster. The method comprises: setting up a simulated application cluster to be tested, the simulated application cluster comprising at least two testing machines; performing an overall workload pressure testing on the test machines of the application cluster to obtain an average single-machine capability of the test machines; and calculating the application cluster capacity based on a total workload pressure of the application cluster and the average single-machine capability of the test machines. By setting up a simulated application cluster to be tested and performing an overall workload pressure test on multiple testing machines of the whole simulated application cluster, the disclosed embodiments can obtain a more accurate average single-machine capability of the machines that is closer to reality, thereby obtaining a more accurate capacity of the application cluster.