Method and System for Dynamically Determining a Power Cap for Individual Servers in a Cluster of Servers
Publication Date: 2010-Jul-21
The IP.com Prior Art Database
A method and system is disclosed for dynamically determining a power cap for individual servers in a cluster of servers. Power capping enables planning and maintaining power requirements of the cluster of servers within a power budget for the cluster of servers.
Method and System for Dynamically Determining a Power Cap for Individual Servers in
a Cluster of Servers
Disclosed is method and system for dynamically determining a power cap for individual servers in a cluster of servers. Power capping enables planning and maintaining power requirements of the cluster of servers within a power budget for the cluster of servers.
In accordance with the method and system disclosed herein, power is distributed across servers in the cluster of servers by defining hierarchy in the cluster of servers. In an embodiment, the hierarchy may be defined based on a Physical Layer, which enables a physical power distribution. In the hierarchy, nodes represent circuit breaker panels, PDUs, and individual servers as leaf nodes as shown in fig.1.
Each layer in the hierarchy is a physical distribution point with built-in software for controlling power demand and distribution for the layer. Further, the software enables various layers in the hierarchy to communicate with each other.
The system employs various policies for defining power cap for each leaf node in the hierarchy and for distributing power to the leaf nodes. The cluster of servers in entirety also has a system power cap. Therefore, the system power cap is one of the constraints which defines individual power cap for each node. In an embodiment, a
power cap for a leaf node may be defined based on workloads and priority associated with each leaf node. In an exemplary embodiment, policies such as waterfilling , Equalperf, Highperf and Maxperf are used for distributing power to each node as per workloads and priority of the nodes. The fig. 2 illustrates various leaf nodes associated with various policies based on their workloads and priority.
At each node, a policy establishes how power from parent nodes is distributed to child nodes. Further, as illustrated in fig.2, virtual nodes are inserted into the hierarchy to represent groups of nodes using same policy.
Each node comm...