Cluster Power Management via Application Migration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current Distributed Power Management (DPM) solutions experience high application performance penalties due to the delay in powering on hosts and redistributing virtual machines, as they require activating host components and reloading the operating system, which is inefficient compared to Host Power Management (HPM) that focuses on single hosts.
Innovation Solution
A method and system for managing power consumption in a clustered computing system by detecting underutilized hosts, migrating applications to more utilized hosts, and maintaining underutilized hosts in a reduced power state without powering them down completely, thereby reducing latency when returning to active mode.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If DPM redistributes VMs among physical hosts by powering on previously powered off hosts, then power management capability across the cluster is improved, but application performance deteriorates due to host activation delay
Solution Approach 1:
The system performs preliminary actions by migrating applications to target hosts before powering off source hosts. This ensures that when the source host needs to be activated again, the application is already positioned on a different host, minimizing the performance impact of host activation delays.
Solution Approach 2:
The system introduces an intermediary management layer that coordinates application migration and host power state transitions. This intermediary manages the redistribution of VMs across hosts, enabling power management decisions to be made without directly impacting application performance through careful orchestration.
2Loss of energy
If DPM powers off hosts to save power, then energy consumption is reduced, but time is lost due to host activation delay when hosts need to be brought back online
Solution Approach 1:
The system performs preliminary application migration before host power-down, so that when hosts are powered off for energy savings, applications are already relocated. This preliminary action eliminates the need for application redistribution when hosts are activated, reducing the time penalty associated with host power state transitions.
Solution Approach 2:
The system changes the operational parameters of hosts by transitioning them between powered-on and powered-off states based on cluster utilization. By carefully managing these parameter changes and pre-positioning applications, the system achieves energy savings while minimizing the time impact of host activation delays.
3Loss of energy
If HPM uses processor ACPI states to reduce power consumption on a single host, then power savings are achieved with minimal latency, but cluster-wide power management capability is lost
Solution Approach 1:
The system merges the benefits of HPM (minimal latency through ACPI state transitions) with the cluster-wide management capabilities of DPM. By combining application migration functionality with host power state management across the entire cluster, the system achieves both energy efficiency and cluster-wide adaptability.
Solution Approach 2:
The system creates a universal power management solution that can operate at both the individual host level (like HPM) and the cluster level (like DPM). The multi-functional system can manage power consumption on single hosts while simultaneously providing cluster-wide load balancing and resource optimization.
Data Source
AI summary
Power consumption by a first host included in a plurality of hosts in a clustered computing system, where each of the hosts executes one or more applications, is managed by detecting that a utilization level of the first host falls below a threshold value and, responsive to the detecting, migrating one or more applications executing on the first host to a second host. After the migration is completed, the first host is caused to consume less power while remaining powered on.