Browse Prior Art Database

Method of Executing High-Utilization Distributed Workloads to Determine Maximum Power Consumption on Interconnected Computer Systems

IP.com Disclosure Number: IPCOM000202538D
Publication Date: 2010-Dec-21
Document File: 5 page(s) / 29K

Publishing Venue

The IP.com Prior Art Database

Abstract

Disclosed is a method for consuming maximum system power on a collection of interconnected power-metered hardware resources by means of maximum utilization system workloads that coordinate all available hardware resources to execute a series of loads targeting smaller subsets of the power-metered hardware.

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 21% of the total text.

Page 01 of 5

Method of Executing High-Utilization Distributed Workloads to Determine Maximum Power Consumption on Interconnected Computer Systems

As the hardware infrastructures of datacenters become increasingly dense through server consolidation and virtualization, the raw power consumption per square foot of server hardware correspondingly increases to levels that are sometimes not sustainable. The burden of additional cooling equipment, which is needed to keep such concentrated high-power hardware running within normal operating conditions, further adds to the power consumption problem. Power budgeting of server hardware to minimize and predict power consumption is therefore critical in containing datacenter costs and maintaining the integrity of the underlying electrical infrastructure.

Datacenter power budgeting guarantees predefined limits on the amount of power that any given server may consume. Such known limits can be used to predict ongoing datacenter costs and plan for future hardware upgrades by determining whether the electrical infrastructure can support the additional predicted load. In order for this planning and cost analysis to be most efficient, the predicted power limits of hardware assets must be highly accurate.

A hardware asset's guaranteed power limit, which is essentially the estimated or predicted maximum power that the asset will consume during its operation, has historically been determined in several ways. Early maximum power estimates utilized the static nameplate power values printed on the labels of the power supplies. These estimates were guaranteed not to be exceeded, but they were overly-conservative and did not accurately represent the actual power consumed by real-world applications. Power budgets based on these conservative values resulted in power being reserved but not used for existing hardware, while new hardware could not be easily added to the datacenter without violating the inflated power budget constraints.

U.S. patent number 5964879 ("Method and system for dynamically power budgeting with device specific characterization of power consumption using device driver programs") describes a method for obtaining a more accurate power estimate by executing a device driver code to vary the power states of devices in the system and measuring the increase in total system power effected by these devices. Such a real-time power measurement scheme provides an evaluation of potential maximum system power consumption that is more accurate than static label power values. However, the invention's device power evaluation excludes the Central Processing Units and does not focus on the overall system power consumption, so the predicted maximum system power consumption must still be partially estimated.

A more accurate method for evaluating and predicting maximum system power consumption is detailed in U.S. patent number 7739548 ("Determining actual power consumption for system power performance states"). The patent...