Transcript Online Grid Replication Optimizers to Improve System Reliability

Green Computing
Green Computing
• Current system extremely wasteful
– Need energy to power
– Need energy to cool
• 1000 racks, 25,000 sq ft, 10MW for computing, 5 mw to
dissipate heat
• Need a system more efficient, less expensive
strategy with immediate impact on energy
consumption
Data Centers
• Focus by green computing movement on data
centers (SUVs of the tech world)
• 6,000 data centers in US
– Consume 61B kWh of energy in 2006
– Cost: $4.5 B (more than used by all color TVs in
US)
– In 2007, DOE reports data centers 1.5% of all
electricity in US
– Greenhouse gas emission projected to more than
double from 2007 to 2020
Data Centers
• By 2012 cost of power for data center
expected to exceed cost of original capital
investment
Goal
• Fed. Gov. wanted data center energy
consumption to be reduced by at least 10% by
2011
– Same as energy consumed by 1M average US
households
Future Vision
• Sources of computing power in remote server
warehouses
• Located near renewable energy sources –
wind, solar
• Usage shifts across globe depending on where
energy most abundant
Current approaches
• Some “low hanging fruit” approaches
– Orient racks of servers to exhaust in a uniform
direction
• Higher fruit - Microsoft
– Built near hydroelectric power in WA
– Built in Ireland - can air cool, 50% more energy
efficient
– Countries with favorable climates: Canada,
Finland, Sweden and Switzerland
Current approaches
• Google – trying to reduce carbon footprint
Carbon footprint includes direct fuel use, purchased
electricity and business travel, employee commuting,
construction, server manufacturing
– According to Google, its data centers use ½
industry’s average amount of power
– How? Ultra efficient evaporative cooling
(customized)
• Yahoo (what is Yahoo??)
– Data centers also carbon-neutral because of use
of carbon offsets
Current approaches
• US government
– EPA has phase-one of Energy Star standards for
servers
– Measure server power supply efficiency and
energy consumption while idle
– Must also measure energy use at peak demand
• Green Grid consortium
– Dell, IBM, Sun VM-Wear AMD
• Green500 – 500 most green supercomputers
Current approaches
• Replace old computers with new more energyefficient
• But manufacturing through day-to-day uses energy
• Dell - reducing hazardous substances in computers,
OptiPlex 50% more energy efficient
• HP – “Greenest computer ever” rp5700 desktop PC
– Died??
• Is MacBook air greenest?
Goals for Future
1. Consider energy to manufacture, operate,
dispose of
2. Sense (?) and optimize world around us
3. Predict and respond to future events by
modeling behavior (grown in performance)
4. Benefit of digital alternative to physical
activities
– E-newspapers, online shopping
• Personal energy meter??
Green Introspection by K.
Cameron
History of Green
• In the 1970s
– Energy crisis
– High gas prices
– Fuel shortages
– Pollution
• Education and action
– Environmental activism
– Energy awareness and conservation
– Technological innovation
Gifts from the 70s
• Energy crisis subsided
• In the meantime advances in computing
responsible for:
– Innovation for energy-efficient buildings and cars
– Identified causes and effects of global climate
change
– Grassroots activism, distributing info about energy
consumption, carbon emission, etc.
• The same computing technologies pioneered
by hippie geeks (???) are the problem now
What happened next
• Call to action within IT community (what
about the 80s??)
• In 1990s
– General-purpose microprocessors built for
performance
– Competing processors
• ever-increasing clock rates and transistor densities
• fast processing power and exponentially increasing
power consumption
– Power wall at 130 watts
– Power is a design constraint
Better, but also worse?
• To reduce power consumption
– Multicore architectures – higher performance,
lower power budgets
• But
– Users expect performance doubling every 2 years
– Developers must harness parallelism of multicore
architectures
– Power problems ubiquitous – energy-aware
design needed at all levels
More problems
– Memory architectures consume significant
amounts of power
– Need energy-aware design at systems level
• Disks, boards, fans switches, peripherals
– Maintain quality of computing devices, decrease
environmental footprint
– Can’t rely on nonrenewable resources or toxic
ingredients
Those data centers
• IT helping in data centers
– Reducing energy with virtualization and
consolidation
– Need to address chip level device to
heating/cooling of building
• Need metrics
Yet another group
• Metrics
– SPECPowerjbb benchmark and DCiE from Green
Grid
• Green Grid – group of IT professionals
– Power Usage Effectiveness PUE
PUE = Total facility power/IT equipment power
– Data Center infrastructure Efficiency metric DCiE
1/PUE
• Benchmark acceptance takes time
Big government
• US EPA Energy Star specification for servers
• Will have impact
– US gov. procurements required to purchase
energy star machines (already true of monitors0
• May be further gov. regulations (with Dems in
power ??)
• EU implemented carbon cap and trade
scheme, US to follow
Trade-off
• How often to replace aging systems?
– 2% of solid waste comes from consumer
electronic components
– E-waste fastest growing component of waste
stream
– In US 130,000 computers thrown away daily and
100 M cell phones annually
• Recycle e-waste (good luck)
• Use computers as long as possible?
The Case for EnergyProportional Computing
Barroso and Holzle (Google)
Intro
• Energy proportional computing primary
design goal for servers
• Cooling and provisioning proportional to
average energy servers consume
• Energy efficiency benefits all components
• Computer energy consumption lowered if:
– Adopt high-efficiency power supplied
– Use power saving features already in equipment
Intro
• More efficient CPUs on chips based on
multiprocessing has helped
• But, higher performance means increased
energy usage
Laptops vs. Servers
• Mobile device techniques
– Multiple voltage planes, energy efficient circuit
techniques, clock gating, dynamic voltage
frequency scaling
– Mobile high performance, short time followed by
long idle interval
– High energy efficiency at peak performance, low
energy inactive states
Servers
• Servers
– Rarely completely idle
– Seldom operate at maximum
– 10-50% of max utilization levels
– 100% utilization not acceptable for meeting
throughput, etc. – no slack time
Servers
– Completely idle server waste of capital
• Difficult to idle subset of servers
– Servers need to be available
• Perform background tasks
• Move data around
• Can help recovery of crash
– Applications can be restructured to create idle
intervals
• Difficult, hard to maintain
– Devices with highest energy savings, highest
wake-up penalty, e.g disk spin up
Energy Efficiency at varying utilization
levels
• Utilization – measure of performance
normalized to performance at peak loads
• Energy efficient server still consumes ½ power
when doing almost no work
• Power efficiency – utilization/power value
• Peak energy efficiency occurs at peak
utilization and drops as util. decreases
• At 20-30% utilization, efficiency drops to less
than ½ at peak performance
Toward energy-proportional machines
• Mismatch between servers’ high-energy
efficiency characteristics and behavior
• Designers need to address this
• Design machines that consume energy in
proportion to amount of work performed
– No power when idle (easy)
– Nearly no power when little work (harder)
– More as activity increases (even harder)
CPU power
• Fraction of total server power consumed by
CPU changed since 2005
• CPU no longer dominates power at peak
usage, trend will continue
– Even less when idle
• Processors close to energy-proportional
– Consume < 1/3 power at low activity (70% of
peak)
– Power range less for other components
• < 50% for DRAM, 25% for disk drives, 15% for network
switches
Dynamic range
• Processors can run at lower voltage frequency mode
without impacting performance
• No other components with such modes
– Only inactive modes in DRAM and disks
• Inactive to active mode transition penalty (even it
only idle to submilliseconds)
• Servers with 90% dynamic range could cut energy by
½ in data centers
• Lower peak power by 30%
• Energy proportional hardware reduce need for
power management software
Inactive/active
• Penalty for transition to active from inactive
state makes it less useful
– Disk penalty 1000 higher for spin up than regular
access latency
– Only useful if idle for several minutes (rarely
occurs)
– More beneficial to have smaller penalty even if
higher energy levels
– Active energy savings schemes are useful even if
higher penalty to transition because in low energy
mode for longer periods
Conclusions
• CPUS already exhibit energy proportional
profiles, other components less so
• Need significant improvements in memory
and disk subsystems
– Such systems responsible for larger fraction of
energy usage
• Need energy efficient benchmark developers
to report measurements at nonpeak levels for
complete picture