liam newcombe bcs data centre specialist group secretary modelling data centre energy efficiency and...
TRANSCRIPT
Liam NewcombeBCS Data Centre Specialist Group Secretary
Modelling Data Centre Energy Efficiency and Cost
BCS Data Centre Simulator
Background
Overview of Tools
Structure of the BCS Simulator
Input Data
Sample Output
Development Path
Background
Power Loss Chain – Fossil Fuel – CPU Used
90% 5% 5%
Transmission Losses
Transformer Losses
Data Centre
40% 25% 35%Cooling Losses
Power Infrastructure
IT Equipment
0.5% of Fossil Fuel Energy
65% 20% 15%Network Equipment
Storage Equipment
Servers
30% 45% 25%Power Supply
Other Components
CPU
20% 80%Idle Time Power
CPU Load Power
35% 65%Heat Exhausted
Electricity GeneratedFossil Fuel
Data Centre
Equipment
Servers
CPU
CPU Utilisation
“For each £ or Watt my Data Centre consumes what output do
I get?”
vs.
“For each Service I deliver what is the financial and energy
cost?”
Data Centre Cost and Energy
Overview of Tools
Tool Coverage
• Existing tools– Computational Fluid Dynamics– Server benchmarking– Chiller technology comparisons– Carbon ‘calculators’– Device specific comparators
Tool Problem
When the only tool you have
is a hammer everything
looks like a nail1
1) Abraham Maslow
Tool Problem
The data centre is not a set of independent components that can be individually tuned
The data centre is a complex and interdependent system
Only a holistic tool can give useful answers
Ap
plic
atio
nW
ork
load
IT D
evic
e
PD
Us
UP
S
CR
AC
s
Ch
iller
Air
Flo
w
Clim
ate
Po
wer
Co
st
En
erg
y U
se
Car
bo
n
Inte
nsi
ty
BCS Simulator Coverage Areas
Var
iati
on
Ap
plic
atio
nW
ork
loa
d
IT D
evic
e
Tim
e
Wor
klo
ad
PD
Us
Ele
ctric
al L
oad
UP
SE
lect
rical
Loa
d
CR
AC
sT
otal
The
rmal
Los
ses
Ch
iller
Clim
ate
Air
Flo
wT
otal
The
rmal
Los
ses
Clim
ate
Tim
e
Po
wer
Co
stT
ime
En
erg
y U
seE
very
thin
g E
lse
Car
bo
n
Inte
nsi
tyS
ourc
e of
Ene
rgy
Dat
a C
entr
e La
yout
Air
Flo
w
CR
AC
Loa
d
Ene
rgy
Co
st
Loca
tion
BCS Simulator Understands Variability
Var
iati
on
Ap
plic
atio
nW
ork
loa
d
IT D
evic
eW
orkl
oad
Po
wer
Co
st
Server Benchmarking
Var
iati
on
CR
AC
sT
ota
l The
rma
l Los
ses
Ch
iller
Po
wer
Co
st
En
erg
y U
se
CR
AC
Lo
ad
Ene
rgy
Cos
t
Chiller Technology Comparisons
Clim
ate
Loc
atio
n
Var
iati
on
IT D
evic
e
Air
Flo
wD
ata
Cen
tre
Lay
out
CFD Analysis
Framework Tool
One tool to cover the entire system, allowing the simulation of components in a realistic context.
No need to develop more partial tools, you can now contribute to the open, trusted tool instead.
Framework Tool
Designed to facilitate the incorporation of new or enhanced component models and data
Publish performance data in standard formats that we can use to compare devices
Structure of the BCS Simulator
BCS Data Centre Model
IT Workload
IT Device Load to Power Function
Data Centre Power Transfer Function
Server Workloads
0 20 40 60 80 100
Percent Load
Pro
po
rtio
n
Server Power Utilisation by Load
0
100
200
300
400
500
0% 20% 40% 60% 80% 100%
Load
Wa
tts
Data Centre Power Transfer Functions
0200400600800
1,0001,2001,4001,6001,800
0 200 400 600 800 1,000
kW
IT Electrical Load (kW)
Inp
ut
Po
wer
IT DeviceData
Overall Energy Use
Overall Costs
IT Device
WorkloadData
IT DeviceData
Workload to Power function
Overall Energy Use
Overall Costs
IT Device Costs
IT Device Costs
WorkloadData
IT DeviceData
Workload to Power function PDU
CRAC Units
Chiller Plant
UPS
Master PDU
TransformerOverall
Energy Use
Overall Costs
IT Device Costs
IT Device in a Data Centre
WorkloadData
IT DeviceData
Workload to Power function PDU
CRAC Units
Chiller Plant
UPS
Master PDU
Transformer
Climate Data
IT Electrical Loads
Overall Energy Use
Overall Costs
IT Device Costs
IT Device in a Data Centre
WorkloadData
IT DeviceData
Workload to Power function PDU
CRAC Units
Chiller Plant
UPS
Master PDU
Transformer
Device Energy Costs
Energy Costs
Allocation Mechanism
Climate Data
Simulation Options
IT Electrical Loads
Overall Energy Use
Overall Costs
IT Device Costs
Energy Costs
WorkloadData
IT DeviceData
Workload to Power function PDU
CRAC Units
Chiller Plant
UPS
Master PDU
Transformer
Power Provisioning
Facility Capital Costs
Facility Costs
Device Energy Costs
Energy Costs
Allocation Mechanism
Allocation Mechanism
Climate Data
Simulation Options
IT Electrical Loads
Overall Energy Use
Overall Costs
IT Device Costs
Per Device Costs
Input Data
Fully Representative Data
15
913
1721
1 7 13 19 25 31 37 43
£0.00
£0.05
£0.10
£0.15
£0.20
£0.25
kWh Unit Cost
Hour of Day
Month
Utility Power Costs
£0.20-£0.25
£0.15-£0.20
£0.10-£0.15
£0.05-£0.10
£0.00-£0.05
Open XML Formats
Sample Output
Example Scenario
Simulator Output
Same Computing Workload
100 One App Per Server Servers
15 Virtualised Servers
10 HPC Grid Servers
Old N+1 Data Centre, Nameplate Provisioning
Simulator OutputOld N+1 Data Centre, Nameplate
Overall IT Device(s) Cost
£0
£200,000
£400,000
£600,000
£800,000
£1,000,000
One App Per Server Virtualised HPC Grid
Co
st
Fixed Overhead PowerCost
Proportional OverheadPower Cost
IT Device PSU PowerCost
M&E Infrastructure,Overprovisioning Cost
M&E Infrastructure,Utilised Cost
Maintenance Cost
Hardware Cost
Old N+1 Data Centre, Nameplate
Overall IT Device(s) Energy Use
0
500,000
1,000,000
1,500,000
2,000,000
2,500,000
One App Per Server Virtualised HPC Grid
En
erg
y U
se
d k
Wh
Proportional OverheadPower
IT Device PSU Power
Fixed Overhead Power
Simulator Output
Example Scenario
Same Computing Workload
100 One App Per Server Servers
15 Virtualised Servers
10 HPC Grid Servers
New N+1 Data Centre, Free Cooling
Simulator Output
New N+1 Data Centre, Free Cooling 19°C
0%25%
50%75%
100%
-100
1020
3040
0.00
0.20
0.40
0.60
0.80
DCiE
IT Electrical Load
Temp
DCiE by IT Electrical Load and External Temperature for Free Cooling
0.60-0.80
0.40-0.60
0.20-0.40
0.00-0.20
Simulator Output
New N+1 Data Centre, Free Cooling 25°C
0%25%
50%75%
100%
-100
1020
3040
0.00
0.20
0.40
0.60
0.80
DCiE
IT Electrical Load
Temp
DCiE by IT Electrical Load and External Temperature for Free Cooling
0.60-0.80
0.40-0.60
0.20-0.40
0.00-0.20
Simulator Output
New N+1 Data Centre, Free Cooling 25°C
Overall IT Device(s) Cost
£0
£200,000
£400,000
£600,000
£800,000
£1,000,000
One App Per Server Virtualised HPC Grid
Co
st
Fixed Overhead PowerCost
Proportional OverheadPower Cost
IT Device PSU PowerCost
M&E Infrastructure,Overprovisioning Cost
M&E Infrastructure,Utilised Cost
Maintenance Cost
Hardware Cost
Simulator Output
New N+1 Data Centre, Free Cooling 25°C
Overall IT Device(s) Energy Use
0
500,000
1,000,000
1,500,000
2,000,000
2,500,000
One App Per Server Virtualised HPC Grid
En
erg
y U
se
d k
Wh
Proportional OverheadPower
IT Device PSU Power
Fixed Overhead Power
Simulator Output
Development Path
Development Path
Data Centre energy use and cost is a complex problem that we are only starting to understand
Deliver marginal and fair share, per business service energy and financial costs to enable holistic carbon management
Development Path
Individual component ‘calculators’ are ineffective without the context of a holistic model
Device level A-G style simple labelling is not effective in a data centre environment
Development Path
? ? ? ? E=mc2
As our understanding of the data centre grows we will be able to use the more complex capabilities of the simulator
Development Path
Opportunities for Further Research and Development
• IT Device Performance– Configuration– Workload– Generalised Metrics ( Performance / Energy )– Networking– Storage
Development Path
Opportunities for Further Research and Development
• Application Performance– Hardware interaction– Impact of the working data set– Optimisation
Development Path
Opportunities for Further Research and Development
• Economics ( Cost and Energy )– Intelligent task deployment– Per task accounting– Automated in-source / out-source decisions– Real time cost management and reporting– Policy based management
Development Path
Opportunities for Further Research and Development
• Physics– Integrated Computational Fluid Dynamics– Chiller plant behaviour and management– Workload assignment
Development Path
Opportunities for Further Research and Development
• Electrical / Electronic Engineering– Power infrastructure performance and loss
characteristics– Dynamic self provisioning electrical
infrastructure