Browse Prior Art Database

Intelligent Management of Server Sleep States in a Cluster Environment

IP.com Disclosure Number: IPCOM000235942D
Publication Date: 2014-Mar-31
Document File: 3 page(s) / 112K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a power-saving feature for a server in a cluster configuration. If the server is not used for a specific period of time, then the feature intelligently powers-off the physical server rather than leave it in a sleep state to continue consuming power.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 51% of the total text.

Page 01 of 3

Intelligent Management of Server Sleep States in a Cluster Environment

Servers in a cluster environment can go into a sleep state if not accessed for a certain period of time. Once in the sleep state, the server remains in the sleep state until the application scheduler wants to allocate that server to execute a job. It goes into the sleep state rather than being powered off because the recovery time to restart the physical server in the sleep state is less that it would be if the server were powered off. However, many times the servers remain in the sleep state for extremely long periods, wasting power. For large cluster configurations, the power waste can be great. A method is needed to intelligently power down the servers in the sleep state to save additional power.

The novel contribution is an additional power-saving feature for powering off a server in a cluster configuration if it is not used for a specific period of time.

To be able to intelligently power off the physical server in the cluster configuration, the base board management controller communicates with the job scheduler. The job scheduler then reviews the jobs in its queue to determine if, in the very near future, the physical server needs to be powered on to execute this job. If so, then the server remains in the sleep state. If not, then this scheduler sends a command to place that physical server, now in the sleep state, into the powered off state.

For clusters of a certain size, when jobs are not being executed, the idle servers go into the sleep state. This occurs in certain cluster configurations today. After a certain period of time, the servers in the sleep state are powered down by the job scheduler, not autonomously. However, based on the size of the cluster,...