|
Multi-Core Computing
Multi-cores mainstream
The power consumed by a server over its lifetime is likely
to cost as much as the server itself if energy costs continue to rise. This
has led to the development of energy-efficient multi-core server processors
says Neeraj Gandhi.
The
face of the data center is changing and it is doing so at a breakneck pace.
Bigger applications and processing needs are forcing businesses to arm their
data centers with denser technologies such as blade servers, faster and efficient
processors, high capacity storage devices and high-bandwidth routers and switches
to transport traffic between the servers and the outside world. All this is
being done to enable data centers to run applications which handle the core
business and operational data of any organization today.
The massive influx of technologies is creating a complex data center that is
costly to manage, consumes more power, and is putting a strain on the infrastructure.
As per the US Environmental Protection Agency (EPA) which observed that data
centers consumed 61 billion kilowatt-hours (kWh) in 2006; thats roughly
1.6 percent of total US electricity consumption, or about $4.5 billion in electricity
costs. Some large data centers draw as much energy as a small town. Assuming
the trend continues, by 2011, the energy consumption by servers and data centers
is expected to nearly double.
This astronomical increase in power consumption and the associated costs to
cool servers in the data center is putting tremendous pressure not only on the
equipment but on the overall revenue of the business as well. The cost of managing
the IT infrastructure (largely servers in the data center) has become the largest
and fastest growing component of overall IT spending for many organizations.
In fact power and cooling are the two most serious issues troubling CIOs today.
Escalating energy cost are only adding fuel to a fire that is already burning.
Arnab Roy, General Manager-Marketing, Sun Microsystems India Pvt Ltd., said,
With a billion people participating online today, the network consumes
more than 100 billion kilowatts of electricity and costs businesses around $7.2
billion in utility bills annually. And since the power consumption of data centers
doubled between 2000 and 2005, it is no surprise that 25 percent of the IT budget
is consumed by energy costs alone. Some analysts have said that infrastructure
power usage will soon cost more than the hardware itself.
In this scenario, multi-core processors are doing their bit to bring down energy
consumption in the data center.
- Energy efficient as it helps eliminate
power and cooling issues
- Simplifies the overall computing infrastructure
requirements
- Reduces costs
- Increases speed and performance which
is important when companies are looking at virtualization in the data
center; the added speed translates into greater available computing
power for enterprise applications running in virtualized environments
- Flexible: clients can decide what resources
to share, what resources to dedicate and can even split resources and
host-shared and dedicated environments on the same system
|
The rationale for multi-core processors
As per Gartner, Inc the number of cores on a chip will double approximately
every 18 months through 2015. This has led to the development of energy-efficient
multi-core server processors. According to Vamsi Krishna, Senior Manager-Technical,
AMD India, Enterprises today face a diverse set of direct and indirect
challenges, while considering power requirements for their data centers. Poorly
designed servers are not power efficient and this results in the consumption
of unwanted power leading to wastage of energy.
According to Morgan Stanley, the energy used to power and cool todays
data centers represents 44 percent of a data centers total cost of ownership;
thats a huge bill for a company of any size. Looking at the flipside,
it can also be a huge source of savings. Jyothi Satyanathan, Vice President,
Systems p, IBM India-South Asia added, Data centers typically consume
15 times more energy per square foot and energy costs are at least 10 to 30
times that of a typical office building and, in some cases, maybe 100 times
higherthats significant power consumption.
Enterprises cannot control the rising cost of energy, as
it is an external entity. However, they can control and conserve the power consumed
in their data centers by choosing energy-efficient servers and components. It
is here that multi-core server processors have a starring role offering, as
they do, raw performance and lower power consumption. To achieve this, they
are more efficient in simultaneous processing of multiple tasks. These processors
have the potential to give data center hardware, as well as its energy savingsa
substantial boost. Multi-core processors can improve system efficiency and application
performance for systems running multiple applications at the same time, thereby
increasing the overall efficiency of a data center.
|
"Enterprises
today face a diverse set of direct and indirect challenges, while considering
power requirements for the data center. Poorly designed servers are not
power efficient and consume [excessive amounts of] power"
-Vamsi Krishna
Senior Manager-Technical,
AMD India
|
|
"Data
centers typically consume 15 times more energy per square foot and energy
costs are at least 10 to 30 times that of a typical office building and,
in some cases, maybe 100 times more energy intensive - thats significant
power consumption"
- Jyothi Satyanathan
Vice President, Systems P,
IBM India-South Asia
|
|
"It
is no surprise that 25 percent of the IT budget is consumed by energy
costs alone. Some analysts have already said that infrastructure power
usage will soon cost more than the hardware itself"
- Arnab Roy
General Manager-Marketing,
Sun Microsystems India Pvt Ltd
|
Organizations are looking at reducing the total cost of ownership. They
are trying to balance their IT budget for driving new initiatives versus the
bulk of their budget supporting maintenance and operations. Industry reports
from Gartner, Inc said that approximately 70 percent of the budget is going
towards maintenance and operations. There is a clear need to reduce these costs,
and power and energy would be key areas that organizations need to focus upon,
said R Ravichandran, Director, Sales (South Asia), Intel Technology India Pvt.
Ltd.
According to Calvin Nicholson, Marketing Manager, Server Technology Inc., the
biggest concern in a data center is the cost to operate devices vis-à-vis
the work taken out of them. He added, The cost of installation, operations
and maintenance are approaching the cost of the server. As power and cooling
costs continue to rise, the situation will get worse, he warned.
Data center energy costs are expected to increase as companies deploy a greater
number of servers, which consume more power, and in the process, emit more heat
that needs to be dissipated. If the current rates of growth continue and
data center efficiencies remained unchanged, data center electric bills and
power requirements will double in less than ten years, and electricity bills
will increase by an additional $200 to $300 million each year, said Satyanathan.
Now that the cost and availability of power is going up around the world
and the cost of a server is approaching the cost to install and operate it for
a year, companies are really starting to look at where they can save money and
be more efficient, added Nicholson.
- The challenge: Evaluating a new
server for your data center is no longer simply a matter of measuring
raw performance. With todays increasing demands, you also need
to consider how much power, air conditioning and space a server consumes.
While traditional metrics are good for calculating throughput, they
dont consider these new power and space demands in the equation.
- Whats driving this new challenge?
Buyers and sellers want to participate in new ways. They want to be
connected and access more services and information in many different
ways. With the explosion of wireless devices, voice and data convergence
and the increasing use of Web applications, data centers are under pressure
to deliver more services, transactions and data to more devices. Demand
for these new services is growing exponentially.
Sun created the Space, Watts and Performance (SWaP) metric to address
this need.
- Performance: Using industry-standard
benchmarks.
- Space: Measuring the height of
the server in rack units (RUs).
- Power: Determining
the watts consumed by the system, using data from actual benchmark runs
or vendor site planning guides.
This metric gives you an effective cross-comparison and total
view of a servers overall efficiency. Armed with this information,
you can accurately compare the performance of different servers and
determine which ones deliver the optimum performance for your needs.
SWaP will help you better plan for current and future needs and control
your data center costs. It is the perfect tool for accurately evaluating
horizontally scaled deployments for the delivery of Web and transaction
services.
Measuring performance
Use numbers provided by a recognized benchmark body or actual in-house,
real-world workloads.
Determining power consumption
Use a power meter that records the total watts used by the system
during the test run. Be sure to use the same configuration used to
produce the benchmark results. To avoid inaccurate measurements, it
is important to take the steady-state power measurement that calculates
usage over the duration of the entire run. If you do not have a power
meter, check with your vendor.
Calculating space needs and total cost
Datacenter racks are expensive real estate, filled with an assortment
of servers, switches, communication equipment, storage arrays, wireless
routers, WAN switches, backup power supplies and more. All these devices
compete for available space and contribute to the cost of powering
and cooling the data center. The true economic value of a server is
determined by the performance it delivers per unit size and the power
it consumes. The SWaP metric effectively and accurately projects and
calculates server efficiency in rack dense deployments, which impacts
data center capacity, performance and costs, while providing imperial
proof points for the new generation of servers.
With the SWaP metric youll have a quick and easy way to accurately
predict the efficiency of a server and the impacts of deploying that
server over your project lifecycle. It gives you the freedom to do
more with less by choosing power and space efficient servers that
reduce overhead cost.
It can help save millions of dollars that can be better used to increase
your companys business value and competitive advantage.
IDCs take
Were entering a global energy crisis where companies
are forced to consider how high energy costs and real estate will
effect their revenue, said Vernon Turner, group vice president
and general manager of enterprise computing, International Data Corporation.
Benchmarking the energy efficiency of IT systems can help customers
make better purchasing decisions when considering the trade-off between
the need for greater performance and the rising cost of energy and
real estate.
Source: Sun Microsystems
|
Flavor of the season
Multi-core is certainly the flavor of the season with Xeon, Opteron, UltraSPARC
T1 & T2 and Power all embracing this approach. The multi-core architectures
of server vendors differ and it is difficult to say whose is the best because
just the fact that a processor architecture is multi-core confers some advantages
on it.
Servers consume a lot of energy, and at the same time a lot of energy is spent
to cool a data center. This essentially means that for every watt of power consumed,
another watt is being spent as overhead for cooling. According to Rajesh Dhar,
Country Manager-ISS, Technology Solutions Group, HP India, An ideal power
situation in a data center should be that for every one kWh of power used not
more than one kWh should be used to cool it. But in reality almost two to two-and-a-half
kWh of power is being used for this purpose. The ideal ratio is 1:1.
To address this problem, vendors are offering a portfolio of energy efficient
products that deliver high-performance and sustainable computing solutions.
Intels quad-core Xeon processors operate at 50W (12.5 W per core) versus
the single core Xeon processors that used to consume 110W. The energy-efficient
quad-core Xeon 7300 series supports frequencies up to 2.93 GHz at 130 watts,
several 80-watt processors and a 50-watt version optimized for four socket blades
and high-density rack form factors with a frequency of 1.86 GHz. Ravichandran
said, Intels internal IT department did a consolidation of servers
from single core to quad core and this has helped achieve 66 percent faster
performance, 66 percent lower power consumption, 87 percent reduction in footprint
and savings of approximately $6,000 for every eight servers consolidated into
one.
Sun Microsystems launched the UltraSPARC (US) T1 processor in November 2005
with eight cores; each core was capable of executing four threads32 simultaneous
threads in all. In August 2007 the company introduced the US T2 (Niagara II)
that has eight cores and can handle 64 threads. It blends high performance and
low power consumption with three functions on one piece of silicon: multi-threaded
10 GbE networking, crypto acceleration, and PCI-Express I/O expansion. Roy said,
The power consumed by the US T2 is less than 95 watts (nominal) i.e. less
than two watts per thread.
The IBM POWER6 processor is a 65-nm 64-bit, dual-core processor with 790 million
transistors running at up to 4.7 GHz with eight megabytes of on chip Level 2
cache. The p6 processor doubles the speed of the previous generation,
the p5, while using nearly the same amount of electricity to run and cool it.
This means that customers can use the new processor to either increase their
performance by 100 percent or cut their power consumption virtually in half,
said Satyanathan.
AMDs quad-core Opteron comes with features like PowerNow and CoolCore
to can switch off computing cores when there is no work load for them to crunch.
This even applies at the transistor level when they are not in use, which result
in additional energy conservation. Krishna said, With the Opteron quad-core
we enhanced PowerNow with a feature called Independent Dynamic Core Technology.
This technology goes a step further and allows each core to vary its frequency
depending on the specific performance need of the applications running on it.
This increased flexibility is designed to help reduce the power consumption
and heat dissipation of quad-Core Opterons when compared to its dual-core predecessor.
AMD Opteron Quad-Core
- Dual Dynamic Power Management helps supply independent voltage to
the cores and to the memory controller.
- Independent Dynamic Core Technology helps reduce system power consumption
by reducing clock frequency per core.
IBM Power6
- Voltage/frequency slewing, enables the chip to lower
electricity consumption by up to 50 percent with minimal impact on performance.
- Processor clocks can be dynamically turned off when there is no useful
work to be done and turned back on when there are instructions to be
executed.
Intel Quad-Core Xeon
- Four-socket, 16-core blades that use less energy.
- Several 80-watt processors and a 50-watt version optimized for four
socket blades.
Sun UltraSPARC T2
- Chip Multithreading Technology is capable of executing 64 threads
simultaneously.
- Powered by less than 95 watts (nominal) with less than two watts
per thread.
|
Market developments
According to Forrester Research, servers consume 30 percent or more of a large
data centers total electricity requirement. Servers also generate heat,
the removal of which from the data center via HVAC systems consumes still more
power. So microprocessor manufacturers are pursuing energy efficiency by incorporating
multiple processors on a single chip (multi-core designs) and using advanced
materials and manufacturing processes that reduce a chips power consumption.
Multi-core processing is a growing industry trend as single core processors
are rapidly reaching the physical limits of possible complexity and speed. The
huge growth of computing requirements and the need to manage bigger applications
has also fuelled the need for such processors. Moreover, with data center energy
costs on course to outweighing the cost of the elements inside a facility, CIOs
have started realizing the magnitude of the problem.
With the increased use of applications that require more power and the
issues of energy conservation, there continues to be a growing need to consolidate
data centers and implement server virtualization technologies. Multi-core processors
seem to be the answer to this problem, said Satyanathan.
Organizations are adopting multi-core based servers. In the recent past
companies such as L&T, Bajaj Allianz and UB have already deployed our quad-core
Xeon servers keeping power problems in mind, opined Ravichandran.
Today businesses around the world employ bigger applications and maintain larger
databases. Along with the increased use of these applications comes an increase
in the amount of power needed to support these applications.
A multi-core processor seems to be an apt solution to this problem.
Data centers equipped with multi-core processors fare better. These chips offer
better performance-per-watt than single-core processors. Resource utilization
being the biggest challenge in a data center, these processors helps in consolidating
resources with better utilization than single-core processors. This is precisely
why the industry is changing from single-core to dual-core and to quad-core
and beyond.
neeraj.gandhi@expressindia.com
|