Browse Prior Art Database

Method for reducing power costs of an IT environment through weather and power-cost aware workload scheduling Disclosure Number: IPCOM000198118D
Publication Date: 2010-Jul-26
Document File: 2 page(s) / 23K

Publishing Venue

The Prior Art Database


A significant portion of operational costs for IT data and compute centers is due to energy consumption. The workload and utilization of a compute system impacts its power consumption and heat production significantly. Produced heat translates to cooling power costs. There exists optimization potential when the compute systems are only partially utilized and when workload schedules can be modified to take environmental conditions such as energy cost at specific times and expected weather conditions (cost of cooling at specific times) into account. In this article an approach to reduce energy consumption by scheduling some of a data center's computer workload to run at times during which power is inexpensive and/or the weather conditions are such that cooling is less expensive. Such scheduling is possible as not all workload has to be executed immediately, some tasks can typically be delayed and scheduled to run at a more optimal time. Examples include backup, archiving, conversion, reporting and analysis runs that can be scheduled in advance.

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

Page 1 of 2

Method for reducing power costs of an IT environment through weather and power-cost aware workload scheduling

This article discussed the use of a predictive modelling of weather and power cost over time to pro-actively schedule computer workloads based on environmental factors in a way that reduces operational cost of the IT infrastructure.

The combination of projected power costs, projected weather conditions (based on cost models and external weather prediction) with workload scheduling allows for a greater degree of optimization that what is possible with traditional scheduling.

The optimization system is extensible. It consists of a core optimization and decision engine that is fed from individual prediction modules with projected cost and demand information. A policy driven decision engine combines static optimization and operation rules with the calculated results based on the combination of the output of the prediction modules. The key prediction modules are the following:

1.1 Optimization based on power tariffs taking changing power costs into consideration
Today energy prices and their changes over time are not considered when shaping computer workloads. (e.g. energy is cheaper at low-demand times and/or when cheap energy like wind energy becomes available, more flexible pricing schemes that expose these dynamic and difficult to predict cost fluctuations to the consumer are expected in future). The system discussed consumes static information about power costs (e.g., such as power tariff tables with varying prices based on daytime) as well as dynamic power cost updates that may be made available on short notice from an external source (dynamic pricing by the power provider) and information about power produced locally by the consumer (e.g. through photovoltaic panels and wind energy). It combines these data into a predictive model provides this model to the main optimization and decision engine.

1.2 Optimization of cooling based on cooling demand history, current weather conditions, weather prediction and weather history
Existing cooling optimizations use environmental factors, such as outside air...