I’ve been conducting data center tours for sales prospects and customers for about five years, however, only in about the last year have I been posed repeated questions regarding environmental friendliness and going green. I am personally skeptical – and even a bit cynical - when I hear corporations’ claims of green. Nonetheless, as Aplicor practices environmental friendliness and also shares its practices with customers, I’m going to use this post to show how going green in the data center can save you some money and mother earth at the same time. Due to the specific content, this post is probably only relevant to Aplicor customers or partners who manage their own data centers and may find our experience on this topic helpful.

The Software as a Service (SaaS) delivery model offers a two punch combination to carbon dioxide emission reductions. The first punch is purely from the economies of scale realized from centralized processing and a shared services model. Instead of thousands of customers individually operating thousands of servers and the power hungry facilities to support those servers, the SaaS multi-tenant model centralizes data center operations to use less equipment and a small fraction of the supporting facility costs. When you recognize that supporting facility costs outweigh the cost and emissions production of the servers and related computer equipment you get a handle on how material this savings really is. The second punch to carbon dioxide emissions is born by the data center operator via thoughtful decision making and a desire for improved power consumption, cooling efficiency and equipment density.

A green data center is one which maximizes energy efficiency and minimizes environmental impact. Anybody can make a difference as operational reforms and rethinking the ways you use equipment may have as big an impact as purchasing newer more energy-efficient computing products. Here are some tasks you can do to achieve lower costs, increased power availability and cleaner air.

• First, build your own power consumption baseline. Understand where you consume power and recognize the trends of where power consumption is changing. Then apply costs to each major power consuming device. Be sure to recognize early that IT equipment does not consume most of the power in the data center. According to a 2007 APC study, the typical power usage allocation is as follows:
  
  
• Turn off your mystery servers. Sun estimates that between 8% and 10% of all data center servers have no identifiable function. It’s a fact that it costs more for power to run a server over its lifetime than the procurement cost of the server. The best thing you can do is turn off unnecessary servers and ancillary equipment. If you don’t know a server’s purpose after an initial investigation, turn it off and wait for the help desk call. If you don’t get that help desk call after 90 days or so, remove the server from the power grid.
• Virtualize servers for greater efficiency. It’s been my experience that most IT staff don’t know how underutilized their servers really are. Server virtualization allocates a single server into multiple, smaller virtual servers without multiple increments of power consumption. Tapping into unused processing capacity without drawing additional power delivers near linear savings to the bottom line. While virtualization provides material energy savings, recognize it must be complimented with new centralization tools in order to provide proactive management, real-time visibility and leveling analysis across your system assets. While virtualization facilitates network management in many ways (e.g. you can configure a new VM in minutes as opposed to hours for a physical server), it also adds complexity as the number of virtual assets will exceed the prior number of physical assets, management of the virtual assets requires new skills and tools and the relationships between physical and virtual assets must be thoroughly understood. My experience is that the learning curve is not long and the tools are not expensive, however, like anything else in the data center the processes must be thought-through for implications, scheduled in advance, performed according to change management and incorporated into all other relevant SOPs (security, redundancy, performance, scalability, system fault tolerance, disaster recovery, etc.)
• Don’t forget about storage. Efficiency improvements are often limited to servers, yet storage systems while fewer in number are generally growing their capacity at 50% per year (source: IDC) and are among the largest energy consumers in the data center. Storage equipment is often ideal for improvement as it consumes 13 times more power than server processors and storage assets often have a very low utilization (often less than 25%). Storage virtualization can drive use rates from 25% to 50% or more. I’ve also discovered energy benefits from inter-operable devices such as hot-swappable disk drives, power supplies and fans  as well as modular components such as blades, power supplies and network connectors.
• Consider network device virtualization. I’ve been reading articles and talking to some data center veterans on this savings opportunity, but quite frankly, haven’t really figured out how to achieve a measurable savings without compromising our HA (high availability), N+2 redundancy. I’ll probably post an update to this entry if I make some progress on this topic.
• Consider data deduplication. In addition to being a good software management practice, this reduces redundant data processed and maintained in storage assets. This twist on this practice is the new technology which creates calculated hashes for each data block so if a duplicate hash matches that of another block already being stored the system doesn’t store the redundant block. Unfortunately, I’ve not found a lot of good products for this at our organization. While I have seen some products geared toward file storage, there seems to be a lack of data deduplication products that work with block LUNs (which is where Aplicor stores the most data). Duplication technologies provide an added boost when considering virtualization. Each VM (virtual machine) redeploys the operating system multiple times. Deduplication tools can reduce storage space requirements and leverage data access to data residing in the filer’s cache.
• Consider thin provisioning. This storage management practice lets storage managers limit the allocation of physical storage based on levels and/or need and then automatically add incremental capacity only when it is actually needed. This capability is generally a SAN or iSCI hardware function and actually allocates disk blocks to a given application only when the blocks are actually written. This removes the initial allocation guessing game and over-allocation of storage processing just in case you need it.
• Consider IT charge-backs. The pay-for-use formulas can assist in improving equipment accountability and utilization.
• Don’t bypass supplemental cooling if you’re using high density equipment. I’ve seen this mistake made several times. Many data centers are already well past the cooling capacity provided by raised floors, which is generally about 4kW to 7kW per cabinet. While I’ve heard claims of higher capacity, I’ve not witnessed a raised floor cooling system exhaust more than 7 kilowatts per cabinet. Cabinets with blade servers are often above 12kW – and I’ve seen some at over 30 kW - and should get supplemental, localized cooling.
• Hot/cold isles. I initially took this practice as a given, however, have engaged in a few debates with data center staff who question the savings. Aplicor data centers use hold and cold isles and I have clearly witnessed the energy benefits. You may want to seek other opinion on this to make your own conclusion. Also, if you choose to implement this policy, make sure all tenants in the colo follow the configuration. I’ve been in a few colo’s where the exhaust sides of equipment don’t always face each other which means that one cabinets heated exhaust becomes somebody else’s incoming.
• Consider data center segmentation. Servers and most network appliances operate well at higher ambient air temperatures than storage devices and some other data center equipment. The one caveat to this statement is the 1U servers which in my experience overheat much more quickly than 2U servers. I suspect, but haven't verified, that the 1U servers have insufficient airflow and inefficient fans as they are so small. Because storage equipment requires approximately 15% to 20% cooler air, isolating this equipment into a separate room would permit the ambient air temperature for the rest of the data center to increase to 78 to 80 degrees F and thereby result in a substantial power savings.
• Cabinet glass doors look cool, however, they turn cabinets into ovens. Avoid them.
• Good cable management both in the cabinet and the raised flooring will result in far better airflow and more efficient cooling. Be sure to keep cables neat, labeled and bundled – and don’t stuff the cables into limited space.
• Don’t over-refrigerate the data center. I’ve heard experts indicate that data center ambient air temperature can be as high as 78 degrees F. As we use fibre SANs which require more cooling than the rest of the equipment I’ve not quite reached that benchmark. Aplicor data centers maintain 70 +/- 1 and humidity at 52% +/- 2. I strongly recommend not cooling much below 70 degrees as it adds cost with no associated benefit. I’ve heard that for every degree increase of ambient temperature, there is at least a few corresponding percentage points in cooling systems energy savings. I’ve not measured this myself and I’m sure the correlation is not exactly linear, but it seems to be largely justified.
• Use 208V/30A power to lessen energy costs. Each cabinet which uses higher voltage 220 instead of 110 power reduces power loss through the normal wire and transformation process which permits the hardware to operate more efficiently.
• Don’t forget to use blank plates in your cabinets. This is a very small effort which prevents hot spots and hot air from blowing into cool isles.
• Don’t use perforated tiles on raised floors in a hot isle. I don’t see this much anymore, but I did just see it in a San Jose data center so I’m adding it to my list.  
• Turn the lights off. Just like your mother told you. Aplicor data centers use motion detector lights so convenience is not lost and savings are achieved.
• Ask your hardware vendors for advice. Every reputable hardware and equipment vendor has staff that can provide advice and design assistance for energy savings. We recently reached out to our SAN vendor (EMC) and found them to be very informative and helpful.
• For smaller UPSs, consider upgrading your UPS. UPSs now have inverter circuitry that turns DC power from batteries into AC power for IT equipment using a base level of power irrespective of load.
• For bigger UPSs, consider replacing battery UPSs with flywheels. This is a significant capital investment so for most readers this is more likely something to look for when evaluating data center colo space. Battery UPSs are messy and an environmental adversity. I’ve often witnessed the inherent management weaknesses such as a lack of battery standardization, loose battery maintenance procedures and no well defined schedule for battery replacement. The best battery replacement I’ve seen is flywheels. I first saw these years ago at the NAP of the Americas (in Miami) and I’m now happy to note that are being adopted at an increasing rate. Flywheels completely replace chemical batteries by spinning a wheel which provides temporary power during disturbances until generator power kicks in. Unfortunately, these devices are still expensive, however, they consume less floor space than batteries so may initially make more sense for data centers in high dollar real estate areas.
• Look for incentives. There are more than 100 state and local utility energy conservation programs which provide rebates or incentives for energy efficiency improvements. For example, California’s Pacific Gas & Electric reimburses a portion of the hardware, software and consulting costs, up to $4 million, for server and storage consolidation programs.
• Make sure IT has visibility to the electric bill. I’ve witnessed a number of facilities where the electric bill is approved by the CFO or facilities people without regard to IT participation. IT staff should be accountable for electricity utilization in the data center and incented for operational savings. For this to be most meaningful, data center electrical draw should be separated from the rest of operations (via sensors or installing a separate power meter).
• Watch for future power supply changes. I think the EPA is going to deliver new efficiency ratings on power supplies next year. This could be a big savings. Between 1kWh and 1.5kWh can be saved for every 1kWh saved at the plug (source: EPA). The EPA is working with the Climate Savers Computing Initiative to develop new power supplies which are about 90% more efficient and can reduce greenhouse gas emissions by 54 million tons per year – resulting in an energy cost savings of $5.5 billion.
• I recognize most readers are in a position to upgrade their data centers and not build new centers, however, if capital investments and new construction are in your future, make sure to consider things like catalytic converters on generators, alternative energy supplies such as photovoltaic electrical head pumps and water cooling technologies.

The savings from these suggestions are real and material. As a quick reference example, a single dual-socket, quad server with sufficient memory can replace 30 older, lightly used single processor systems. The power savings are in the range of 12 to 15 kilowatts and could easily translate to $15,000 or more per year.

For another perspective, below are the “Ways To Go Green” results from an Infoworld research project completed in 2007 from a survey of 472 business technology professionals.

Which of these technologies will you use to save energy in the data center?

Data center power growth is a significant and growing problem. According to analyst firm Gartner, 50% of current data centers will have insufficient power and cooling capacity to meet the needs of high density equipment by 2008. A 2007 survey of 369 IT professionals performed by OnStor reported that 63% of respondents have run out of space, power or cooling without warning.

So what are the industry leaders doing?

• As part of its Greenfield project, HP is consolidating 85 global data centers into 6 which will leverage low power/high capacity blade servers, upgraded power and cooling devices and extensive virtualization.
• On August 21, 2007, Sun finished a 12 month construction project which delivered a new 76,000 square foot data center in Santa Clara which uses less than half of the electricity of its previous data centers.
• On August 1, 2007, IBM announced it will consolidate 3,900 servers into 33 virtualized System z mainframes resulting in about 80% power reduction and a $250 million savings in energy, software and system support costs over five years.
• Database vendor Sybase recently reduced the number of servers by about 45%. The company expects to achieve $1 million to $2 million per year in operational savings and has opened up much more space thereby canceling a plan to build out another data center to accommodate growth.

Each of these new data centers will power down idle servers, provision new servers more intelligently and increase storage utilization to the 70 to 90 percent range.

What's the government doing?
The US government has resourced a number of smart folks who provided a macro level data center base line (see www.energystar.gov/datacenters for more information). According to the Department of Energy (DOE) and Environmental Protection Agency’s (EPA) August 2nd report to Congress, US data centers consumed 61 billion kilowatt hours in 2006 – which equates to 1.5% of all electricity consumed in the country at a cost of about $4.5 billion. The DOE forecasts that data center energy consumption will grow 12 percent annually through 2011. So if we do nothing, data center power usage will double by 2011 or if we implement the EPAs recommendations we could lower overall data center power usage to 2001 levels by 2011 – a savings difference of 90 billion kilowatt hours.

On September 18th in New York, IT industry leaders and the 92 member Green Grid Alliance (made up mostly of IT professionals) completed a memorandum of understanding with the DOE that solidifies the process to establish the metrics for data center efficiency measurement. Making the data center the initial focal point was viewed as strategic for two key reasons. First, data centers are extremely large energy consumers, their consumption is rising at an aggressive pace and any offset to this rising consumption may bring a big impact. Second, their operations can be isolated for efficient measurement, modeling and extrapolation. Lessons learned can be shared for broad involvement.

What can you do?
Most important, the memo of understanding sets an objective of using the research and analysis to improve overall data center energy efficiency by 10 percent in 2011. However, putting first steps first, the next action items include the DOE publishing initial metrics specifications (allegedly in December, however, I suspect that date is too aggressive) and the participating community performing a maturation process to settle those metrics throughout calendar year 2008 (my forecasted date only; not theirs). Want to get involved and compare your data center metrics with others? The DOE will be publishing a new web site with the metrics and a suite of tools in the near term. Anybody is free to use the tools and encouraged to submit comments on the metrics. I can assure you I will be performing a comparison analysis and suspect submitting several comments. For example, I’m a believer that corporate America won’t achieve wide spread adoption until it makes financial sense. If the feds were to match state enticements or provide even modest additional financial or tax incentives, I suspect it would tip the balance for many companies to act and benefit the economy for all. Have your own suggestions? I'd like to hear them.

Technorati: Aplicor, CRM     Add to Technorati Favorites
Del.icio.us: CRM, Aplicor      Save this page to del.icio.us
View CC license
Posted by Chuck Schaeffer on October 1, 2007 in Hosted Delivery
Permalink | Comments (0) | TrackBack (0)