Measuring and analyzing the energy use of enterprise computing systems Xiuming Zhang.

Measuring and analyzing the energy use of enterprise computing systems

Xiuming Zhang

System Deployment

• Powernet: a hybrid sensor network that monitors the power and utilization of the IT systems in a large academic building

• Over more than two years• 250+individual computing devices• One power meter per device

Appliance Types Measured

Extrapolating to the Whole Building

Power Share by Type

• 4 types: PC, LCD, server, switch

PC

• Binned into 3 categories– Laptops– Low-end– High-end

• 742 in total– 456: description available. The other: assume same distribution.

LCD

LCD

• Lower brightness & use dark backgrounds

Server

• Monitors 32 of the 500 servers• Calculated average power: 233 W • Estimated total power of 500 servers: 117 kW

Switch

Switch

PC CPU Utilization

• 95% of the computers have a utilization rate lower than 30%

Popular Workloads

Network Switch Utilization

• The demand never exceeded 200 Mbps

Network Switch Utilization

• Highly underutilized• Total network demand < 1000 Mbps 100% of

the time.

Discussions

How does the sampling frequency affect what the data reveal?

• Too small hide the anomalies

How big is the variance between two instances of same model?

Does sampling a few devices provide an accurate average measurement?

shows the wide distribution of desktop power

Does sampling a few devices provide an accurate average measurement?

• Large sample size desired for PC• 1,000,000 random samples of size 5, 10, and 20, drawing from 69 machines

Do short-term measurements accurately reflect long-term power draw?

• One-month scale yields an acceptable error

Are Energy Star data representative?

• Energy Star Standard does not consider PCs under load

Thanks for the attention!