Container CPU Usage Limiting via Network Traffic Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current CPU resource limitation mechanisms, such as cgroups, fail to account for network access CPU usage in container environments, leading to potential CPU and network abuse as they only restrict container process CPU usage without considering network access CPU consumption.
Innovation Solution
The method involves determining network CPU usage by measuring the time spent in software interrupt service routines, assigning weights to packet types, and calculating total container CPU usage, which is then used to adjust the container CPU consumption quota and network bandwidth settings to ensure compliance with predefined limits.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If cgroups are used to limit container CPU usage, then container process CPU usage can be restricted, but network access CPU usage is not accounted for, leading to potential CPU abuse
Solution Approach 1:
The patent segments CPU usage monitoring into two distinct components: process CPU usage (tracked by cgroups) and network access CPU usage (tracked by measuring software interrupt service routine execution time). This segmentation allows each component to be monitored independently using appropriate mechanisms, then combined to provide comprehensive CPU usage accounting that prevents abuse while maintaining system manageability.
2Speed
If network bandwidth is increased for a container, then network performance improves, but CPU usage for packet processing increases, potentially exceeding CPU quotas
Solution Approach 1:
The patent implements a feedback mechanism that continuously monitors network packet processing CPU usage and adjusts network bandwidth allocations accordingly. When network CPU usage approaches or exceeds the container's CPU quota, the system reduces network bandwidth to bring CPU usage back within limits. This feedback loop ensures that network performance improvements do not cause unsustainable CPU consumption.
Solution Approach 2:
The patent dynamically adjusts network bandwidth settings based on real-time CPU usage conditions. The system transitions from static bandwidth allocation to dynamic bandwidth control, where bandwidth is adjusted up or down depending on whether the container is approaching its CPU quota limit. This dynamic adjustment allows the system to optimize network performance when CPU resources are available while preventing CPU abuse when resources are constrained.
Data Source
AI summary
In an approach to limiting container CPU usage based on network traffic, a packet CPU usage for each packet type of one or more packet types is determined. A network CPU usage is calculated for a specific container based on the network CPU usage for each packet type. A container process CPU usage and a total container CPU usage are calculated for the specific container, where the total container CPU usage is the sum of the network CPU usage and the container process CPU usage. Responsive to determining that the total container CPU usage exceeds a threshold, the container CPU consumption quota and the container network bandwidth setting for the specific container are adjusted to reduce the total container CPU usage, using a set of pre-configured parameters.