Computing Service Capacity Management via Predictive Host Scaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The variability in launch times for computing instances on physical hosts due to factors like co-location and resource availability makes it challenging to meet target launch times in virtualization environments, leading to inefficiencies and potential service level agreement (SLA) breaches.
Innovation Solution
A system that uses machine learning models to predict launch times based on current computing capacity and demand, allowing for the addition of physical hosts to ensure that launch times meet target requirements by optimizing co-location and resource allocation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If multiple computing instances are co-located on a single physical host to improve resource utilization, then resource sharing efficiency is improved, but launch time variability increases and target launch times may not be met
Solution Approach 1:
The system performs preliminary actions by pre-identifying suitable physical hosts for computing instance launches before actual deployment. The capacity management service analyzes current host states, predicts launch times using machine learning models, and selects optimal hosts in advance, thereby reducing launch time variability while maintaining resource utilization efficiency through proactive resource allocation
Solution Approach 2:
The system implements feedback mechanisms by continuously monitoring actual launch times and comparing them against target launch times. Machine learning models are trained on historical launch data to predict future launch times, and this feedback loop enables dynamic adjustment of instance placement decisions to meet target launch time requirements while optimizing resource utilization
2Loss of time
If computing capacity is increased by adding physical hosts to reduce launch times, then launch time performance is improved, but system complexity and infrastructure costs increase
Solution Approach 1:
The system changes parameters by dynamically adjusting instance placement decisions based on real-time host state analysis and machine learning predictions. Instead of adding physical hosts, the capacity management service modifies placement parameters to select optimal hosts from existing infrastructure, thereby reducing launch times without increasing infrastructure complexity
Solution Approach 2:
The system uses machine learning models that create virtual copies of host states and launch scenarios to predict outcomes without actual physical expansion. These predictive models allow the system to evaluate multiple placement options and select optimal ones, achieving performance improvements through intelligent software-based decision-making rather than hardware expansion
3Reliability
If instance launch scheduling is optimized to meet target launch times, then service level agreement compliance is improved, but computational overhead for capacity management increases
Solution Approach 1:
The capacity management service implements self-service by autonomously analyzing host states, predicting launch times, and making placement decisions without requiring complex external orchestration. The machine learning models automatically learn from historical data and adapt to changing conditions, enabling the system to maintain SLA compliance through self-directed optimization while managing computational overhead efficiently
Data Source
AI summary
A technology is described for managing computing capacity within a computing service environment. An example method may include determining a state of a computing service environment having a group of physical hosts that execute computing instances and determining a predicted demand for computing instance launches on the group of physical hosts. A co-location of the computing instance launches on the group of physical hosts may then be determined by applying the predicted demand for computing instance launches to the state of the computing service environment. A predicted launch time for the computing instance launches may then be determined using the co-location of the computing instance launches as an input feature. An additional physical host may then be added to the group of physical hosts to decrease the predicted launch time.