Workload Placement Management Using Server Efficiency Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Data centers face increasing power and cooling demands, with existing methods of consolidating workload inefficiently due to the assumption that heavily loaded servers operate efficiently, and the need for suspend/resume capabilities that are rarely implemented, leading to suboptimal energy savings.
Innovation Solution
A method that accesses stored relationships between power consumption efficiency, capacity utilization, and ambient air temperature to calculate current and predicted power consumption efficiencies across servers, identifying the server with the greatest improvement in efficiency and assigning workload to optimize power usage.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of energy
If workload is consolidated from several lightly-loaded servers to one heavily loaded server, then power consumption is reduced, but the assumption that heavily loaded servers operate efficiently is mistaken and suspend/resume capability is rarely implemented
Solution Approach 1:
The patent changes the parameter used for workload placement from simple load balancing to power consumption efficiency, which is determined by multiple factors including server utilization, ambient temperature, and server-specific efficiency characteristics. This allows the system to select servers that minimize power consumption while maintaining operational flexibility, as servers are not simply consolidated into heavily loaded states but are selected based on their efficiency profiles across different operating conditions.
2Productivity
If servers are operated at higher capacity utilization, then fewer servers are needed, but power consumption efficiency does not necessarily improve due to non-linear fan power requirements
Solution Approach 1:
The patent applies local quality by recognizing that different servers have different efficiency characteristics at different utilization levels and temperature conditions. Rather than assuming all servers follow the same efficiency curve, the system uses server-specific power consumption models that account for local variations in fan efficiency, heat dissipation characteristics, and component power consumption patterns. This allows optimization at the individual server level rather than applying a universal utilization threshold.
Solution Approach 2:
The system dynamically adjusts workload placement decisions based on real-time or near-real-time conditions including ambient temperature, current server utilization, and predicted future states. The power consumption efficiency is not a static property but is recalculated as conditions change, allowing the system to adapt to varying thermal environments and load patterns to maintain optimal efficiency.
3Temperature
If fans run faster to remove increased heat from higher capacity utilization, then heat removal increases linearly, but power consumption increases by the cube
Solution Approach 1:
The patent performs preliminary calculation of predicted power consumption efficiency before making workload placement decisions. By using stored power consumption models and current environmental conditions to predict future efficiency states, the system can anticipate the cubic power increase from fan acceleration and avoid placing workloads on servers where this would occur, instead selecting servers that will operate in more efficient power-temperature regimes.
Data Source
AI summary
Additional workloads are assigned among servers in a power-efficient manner. For each of a plurality of servers, a stored power efficiency/capacity utilization relationship is accessed, current component power consumption values are obtained, and a current power consumption efficiency is calculated. An amount of capacity utilization necessary to perform an additional workload is obtained, and a predicted power consumption efficiency is determined for each server. The predicted efficiency is determined using the current power consumption efficiency of the server and the stored relationship. The workload is then assigned to the server that would have the greatest improvement in power consumption efficiency.