Cloud Resource Termination via Exponential Smoothing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
There is a challenge in balancing the timely termination of unused cloud resources to save power and cost while minimizing latency when new tasks are added to a workload, as immediate termination wastes resources and premature retention wastes energy.
Innovation Solution
A controller uses a modified exponential smoothing method to predict task additions, estimating the required cloud resources by adjusting the number of tasks expected in each interval, allowing for delayed termination of unused resources based on the equation Ft=1−αn−1Ft−n+αAt−n−Ft−n, where α is a smoothing factor, to optimize resource management.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If unused cloud resources are terminated immediately to save power and cost, then energy consumption and cost are reduced, but latency increases when new tasks are added
Solution Approach 1:
The system performs preliminary actions by predicting future task additions using exponential smoothing and maintaining cloud resources in advance based on these predictions. The controller estimates the number of tasks that will be added in future time intervals and proactively retains cloud resources to avoid termination, thereby reducing latency when new tasks arrive while still allowing energy savings when predictions indicate no future tasks.
Solution Approach 2:
The system dynamically adjusts cloud resource termination decisions based on predicted task patterns. Instead of static immediate termination, the controller uses exponential smoothing to continuously update predictions about future task additions and adjusts retention decisions accordingly. This dynamic approach allows the system to adapt to changing workload patterns and optimize the balance between energy savings and latency.
2Speed
If cloud resources are maintained to reduce latency, then task addition speed is improved, but energy waste increases
Solution Approach 1:
The system implements feedback mechanisms by continuously monitoring actual task additions and using this information to refine predictions through exponential smoothing. The controller compares predicted task patterns with actual observations and adjusts future predictions accordingly, allowing more accurate determination of when to terminate resources. This feedback loop reduces energy waste by preventing premature retention while ensuring resources are available when needed.
Solution Approach 2:
The system changes the parameter of resource retention decisions from binary (immediate termination or permanent retention) to a dynamic state based on predicted task patterns. The controller uses exponential smoothing to calculate expected future tasks and adjusts retention decisions based on these predictions, allowing resources to be maintained only when predictions indicate future workload, thereby reducing energy waste while maintaining availability when needed.
3Productivity
If exponential smoothing is used to predict task additions, then resource termination is optimized, but processor intensity increases
Solution Approach 1:
The system applies partial action by using a simplified exponential smoothing formula that only processes data from the most recent time interval rather than maintaining comprehensive historical data. The controller uses the formula Ft=Ft-1+αAt-1-Ft-1 which requires only recent past values rather than extensive historical analysis, reducing processor intensity while still providing optimized resource termination decisions based on current workload patterns.
Data Source
Figure 1
Figure 2
AI summary
When one or more tasks require compute/storage resources for their execution, it is well established to provide these resources in a cloud, such that the tasks can be executed on those cloud resources. Typically, when one or more tasks are added to a workload, additional cloud resources are required to be initiated to accommodate this increased workload. Initiating a new cloud resource introduces latency into the process, and so it is desirable to have such additional cloud resources initiated in advance of new tasks being added to a workload. However, keeping unused cloud resources running and unused is wasteful, and therefore it is also desirable to terminate any unused cloud resources as soon as possible after the tasks they are executing have completed. One approach would be to use exponential smoothing to slow the rate at which unused cloud resources are terminated; however, this method is processor intensive. With present invention it is not necessary to track the history of multiple data points from each time interval. Rather, it is only necessary to keep a record of the number of tasks that were added to the workload during the first time interval; that is, the controller only needs to keep a record of time intervals in which new tasks are added to the workload. This is because, in the intervening intervals, in which no new tasks were added, a simplified version of exponential smoothing can be applied.