Improving Data Centers’
Energy Reliability & Efficiency
Background The challenges Alternative solutions SATEC approach Advantages Cases Summary
Data Centers Energy Challenges
Increased demand for On-line Clouding services
Data center energy consumption per m² is ×15-20 times higher compared with typical office
Expected increase in Power consumption
While
Worldwide demand for cleaner power generation solutions Reducing power consumptions by increasing efficiency (PUE)
70%
Type of data centers
Corporate:
In-House Data Center
Collocation:
Hardware (servers), belong to customer
Facility and internal systems maintenance – by provider
Managed/Hosting (“cloud”):
Hardware may be owned by the customer or the provider Hardware can be separated (at rack/ server levels) or shared Provider operates storage, engineering, security, electricity etc.
Tier 4 Most Common Tier 3 Tier 2 Tier 1
Data centers
Reliability & Availability
Increased outsourcing services
Enterprise Corporation 99.995% Uptime
23.3 Minutes Downtime (Y) 2N+1 Fully redundant
96 Hour Power Outage protection
Large Businesses 99.982% Uptime
1.6 Hours Downtime (Y) N+1 Fully redundant
72 Hour Power Outage protection
Medium-size Businesses 99.749% Uptime
22 Hours Downtime (Y)
Partial redundancy in Power and Cooling Small Businesses
99.671% Uptime
28.8 Hours Downtime (Y) No Redundancy
Data Center – Common Metrics
Data Centers Energy Challenges
“Worldwide Data is set to grow 10-fold by 2020 from 4.4 zettabytes to 44ZB” (IDC)
Reliability
Reducing investments Density
Efficiency (PUE)
Improving Reliability
Load Measurement and notifications – down to rack level Power Quality Measurements and alerts, at critical
locations
“what if” testing procedures for measurement of loading
in case of failures
Improving Efficiency
Measurement down to servers racks for accurate PUE
segmentation
Measurement and Control of HVAC systme including
temperature and humidity levels
Improving DC Reliability
and Efficiency
Limited energy information – energy metering at
MCCBs only
Price of Smart PDU prevents wide deployment
Lack of power quality monitoring
SATEC suggested approach
Balancing between expensive Smart PDU and lack of information of board measurement level only
EM720 High performance revenue meter + PQ & Transient PM180 High performance analyzer (PQ) BFM II Energy + Rev. meter
Main Input Backup
ATS UPS UPS UPS HVAC Distribution Boards UPS Switch Gear
~ ~
[N+1] Generators
Real time PDU level load monitoring for loading and
PUE segmentation
High Accuracy energy monitoring
(utility check-meter, collocutors billing, losses)
Cost effectiveness
Power Quality analysis
Leakage current alerts
HVAC optimization based on actual data at rack
level
What-if analysis
Distribution Switchboard
Metering Scheme
Up to 200m
Temperature & humidity (Analog Inputs)
Up to 54 channels (27 racks)
Split/Solid Core High Accuracy Current Sensor
PDU’s
* BFM unit price based on 54 channel BFM II
Price comparison (CAPEX)
PDU Type Av. Unit price Av Rack price (A-B) standard PDU) Delta (from
Regular PDU 200 400 -
Metered PDU 700 1,400 1,000
Smart PDU 1,100 2,200 1,800
Error of Measurement (example)
Meter PDU SATEC BFM
Error for single metering point 2% 0.5% The mean square error of measurement
for 1 rack
(2 metering points)
2.8% 0.7% The mean square error of measurement
for 1 cage
(60 metering points = 30 racks)
15.5% 3.87% Total Power Reading Error
5kW×30 racks×40% loading×%error ±9.3kW ±2.3kW Billing error
(one cage, one year, US$ 0.15/kWh): US$ 12,220 US$ 3,022 Delta charging per cage US$ 9,198
) ( ) ( ) ( ) (x y z U2 x U2 y U2 z U + + = + +
In advance info – to prevent shutdown and downtime
Standard 6 pulse (2 pulse for 1~) rectifiers (UPS,
server PS) cause harmonics. Modern AFE rectifiers reduce harmonics, but increase transients
Transient frequency are high and can pass-through
UPS system
“clean network” is not 100% “clean”
High end Power Quality systems upstream and
downstream of UPS allow measurement of PQ
With advanced thresholds and logs, prevention
action can be conducted and save $$$
Downtime
DC in Russia – lost 20% of servers
Finding – Automatic operation of servers at specific
time (acceleration of computing power) lead to over loading of transformation currents
Solution – constant measurement of racks levels +
logs + thresholds could provide in-advance notification and save
$$$
PQ and Safety Issue
Investments for new DC
=3000-5000 USD/m2
DC (1500 m2)= 3-5M$
PQ+ Fault Recorder+ High Accuracy Billing Meter= 3000 USD
Complete System for Power Quality Control needs 10-12k$ Investments 4-5 measured points
Real time PDU level load monitoring for loading and
PUE segmentation
High Accuracy energy monitoring (billing, losses)
Cost effectiveness
Power Quality analysis
Leakage current alerts
HVAC optimization based on actual data at rack
level
What-if analysis