Posted by Mark Harris on Tue, Jun 07, 2011 @ 03:09 PM
A great thing has happened last month relative to the magical concept of Power Usage Effectiveness (or "PUE" as commonly referred to)... A new version has arrived! Finally we all took a deep breathe and also a NEW look at the concept of performance management of power at the data center system level. PUE Version 2 is born!
Huh? A new version of PUE? How did you miss it? It's not that old, so if you did miss it, read on...
I would say that "PUE" became very newsworthy again to the masses on May 12th (2011) when Matt Stansberry, Director of Content and Publications, Uptime Institute released the findings of a widescale survey on the usage of the original version of "PUE". Several key findings of the survey:
PUE as a performance metric has high adoption rates, but easily can lead to inaccurate reporting and misrepresentation.
Adoption of carbon and water usage reporting metrics (WUE, CUE) will be important to the data center industry globally.
About 55% of all data centers are still cooling at or below 70 degrees F
Only 12% of the sample respondants are managing temperature as efficiently as possible through the use of INLET monitoring and associated control as compared to 40% still using return air temperature (at the CRAC/CRAH)
Thank goodness what we all suspected was finally being printed on paper! It dawned on everyone that the PUE badge of Honor was being used as a declaration of manhood (some call it "PUE envy"), and as such the amazing values reported were being gathered in some extremely creative ways. (If you remember 5 years ago when the PC video card chip makers were all tweaking their designs to assure that their offerings would remain king of the hill when being compared against the competition. They simply tuned their designs for the benchmark audience, rather than real-world usage. Pretty clever huh?). With PUE version 1, I could imagine a data center operator waiting for midnight on a cold winter's night in January to do the calculations, at a time where workload was tiny and most of the chillers were off-line. PUE in this scenerio would look pretty good, No?
Then on May 17th (2011) the Data Center Efficiency Task Force issued a 14 page document entitled “Recommendations for Measuring and Reporting Version 2 – Measuring PUE for Data Centers” – also known as PUE 2. This is one of the most profound changes I have seen in a long time in the area of data center efficiency.
One of the biggest changes? PUE2 has 4 categories, the first one still based upon kilowatts, while the other three categories based upon energy usage, or kilowatt hours. Additionally, the first category, the one based on kilowatts, now requires the kilowatts to be calculation be based on the highest power used by the facility in a 12-month period. Whew, finally we see a set of time based metrics, rather than a best-case snapshot approach.
It should be noted that PUE2 is meant to build upon the tremendous work done years ago by The Green Grid regarding the original PUE metric. Prior to TGG, nobody was looking at the math of deliverying IT. PUE version 1 set the stage for the math! Additionally, PUE2 reflects and reiterates the “Harmonizing Global Metrics for Data Center Energy Efficiency” definition of PUE which was released by the task force in February 2011.
I applaud the DCET and the birth of PUE Version 2.0!
Note: A full rundown was published over at DataCenterKnowledge.com and can be found (in two parts) here:
http://www.datacenterknowledge.com/archives/2011/06/01/power-usage-effectiveness-2-the-sequel/
http://www.datacenterknowledge.com/archives/2011/06/01/pue-2-the-sequel-continued/
Posted by Mark Harris on Wed, Apr 13, 2011 @ 01:40 PM
Funny how a company like FedEx, a heavy industrial behemoth, has jumped to the Data Center Efficiency front row by understanding how technology can be wrangled to support their BIG COMPANY vision, and being highly motivated to DO so. With their IT leadership, they tasked their teams to 'think different' and demonstrated their nible approach at nearly every step of the way in their new 150 throusand square-foot Colorado data center. And the results: IT Efficiency Delivered! PUE 1.28 with FREE cooling for nearly 57% of the time! Now that shows just how efficient a data center can be if you just think it through a bit.(See ).
Their IT master, Kevin Humphries, built the data center in support of his long-term goal, customer effectiveness, corporate efficiency and low impact on the world around them. A good corporate citizen approach as it were. True he did have some help from mother nature by chosing Colorado, since the low humidity and temperatures are ideal for cooling a data center. The talent pool is also very good and power is readily available. Kevin expects to bring FedEx's core processing from Memphis to the new Colorado center over the next few years.
To reach these kind of new PUE numbers, it is clear that investments in new long term strategies need to be developed and deployed. Variable Frequency Drives for cooling, granular asset management and power control and management, aggressive server sprawl and orphan control, economizer cooling, intelligent lighting systems, etc. It all adds up to REAL money that affects the bottom line.
Remember of few years ago when all the rage was Google's proclaimation that they had hit a ground-breaking low PUE of 1.21 and everyone wondered how they did it? Seemed like pure magic must be afoot, but now we see companies like FedEx and others jumping to the head of the class. Big brand-name companies will be strutting their Data Center victories over the coming year or two.... Maybe we should create an award, like the Oscars in entertainment, but for power aficionados, or should we say "efficienados", and we can call it the "Golden Joule" award.
How low can it go? Let's get back to this question in a couple of years, after we see what Intel and AMD have in store for us. After we see the next gen of HyperVisors and solid-state flash memory. We'll see power provisioning and active management of resources like never attempted before. We might even see some of the Autonomic or Gartner's Real-Time Infrastructure re-surface now that technology has caught up...
Posted by Mark Harris on Wed, Mar 23, 2011 @ 03:31 PM
I ran across a great article the other day written by Dan Woods and posted on Forbes' website blog (see note below for URL). In it, Dan starts to articulate some of the dozen or so power-related issues the IT industry as a whole is wrestling with and which should be on every CIO's top of mind. Add to that list a few fundemental business concerns that have been presenting themselves across the various industry forums and summits, and we can start to see a pattern of critically important Energy knowledge that has historically been ignored!
- How much energy do I consume, by location, by project, by organization, etc. ?
- What was my baseline usage last year? what is it today and what would it be at the same growth velocity a year from now?
- How frequently DO I monitor my power usage and how frequently SHOULD I monitor it?
- How 'full' are my company's data centers? Not just physical space, but power and cooling. Can I add a new business app?
- Do I know where my servers are working hard and efficient and conversely do I know where I have orphaned or zombied servers consuming 70% of their maximal operating power and doing nothing?
- Do I know how power is being consumed and where my ROI is for that power and cooling costs?
And lastly...
- Do you have a roadmap for improving power management capabilities?
These are all great questions to ask, and creating the infrastructure to answer these questions is a great way to get your hands around the matter, set a goal or deliverable. Remember, Power is no longer trivial in cost and Data Center power costs will exceed new server acquisition costs in just a year or two from now. They are almost equal today!
Going to the last bullet above is the key. Do we all have a plan? ANY plan? We don't all have to buy the latest and greatest hardware or software, and one size does NOT fit all, but everyone needs to have a Power plan. Your CIO needs it. Your CFO and CEO needs it.
Technologies like the EnergyClous System from Racktivity, can be a real contributing part of that power plan. We see in minute detail everything from the outlet upwards. Physical and logical, enterprise-wide. Consumption and Power Factor, Power Quality and Harmonics. We see it all and can help YOU see it too! (Just talk to us...)
URL for Dan's blog: http://blogs.forbes.com/ciocentral/2010/11/23/power-management-questions-for-every-cio/
Posted by Mark Harris on Tue, Feb 22, 2011 @ 07:40 AM
Ok, here is a site I just have to share with everyone. The folks at copper.org have written a number of stories on Power Quality, which starts with a great Primer on the topic, and by navigating the left-bar stories includes topics like Harmonics and Grounding. Spend 15-20 minutes on their site and you'll feel empowered to intelligently discuss data center design from a power quality perspective.
http://www.copper.org/applications/electrical/pq/primer.html
(While you are over at Copper.org, also browse under the Education Tab and read the details about the Statue of Liberty. Along with all of the obligatory details, there are a ton of historical photos and compromises that were made over the course of the past 150 years or so. Interesting....)
Posted by Wilbert Ingels on Mon, Feb 21, 2011 @ 11:50 AM
John Rath of the Data Center Links blog has recently noted an interesting and useful article in the Black Swan Real Estate eZine that listed the top 10 places you do not want to locate your data center. John's Blog can be found at: http://datacenterlinks.blogspot.com/
The place he cites included:
-
Below sea level or in a floodplain
-
In a location that suffers from frequent natural disasters
-
Near the manufacture, storage or transport of hazardous chemicals
-
In the basement
-
In a building that used to be something else
-
Large metropolitan area
-
In a building with any other business
-
In the flight path of a major airport
-
In a location that has expensive or limited telecommunications
-
Too close to your other one
I am not sure if his above list of places is intended to be a specific ranking but clearly all ten are not suitable locations to locate your data center.
This is just common sense today.
Let's me also add one more place to avoid which is critical for today's new IT strategies: Any place with an unknown or limited availablility of ENOUGH POWER CAPACITY. Remember Data Centers today run dense and hot, and they get bigger over time. Common centers consume a megawatt or more. Even if the metro area has modern power transmission infrastructure, you may find it prohibitive or impossible to fund that 'last mile' of power transmission to the actual data center site being considered. (In the good ole' days, facilities managers made the safe assumption that the power company could bring in as much power as IT could possibility use. There simply was more power outside the building than people could use inside the building. This is no longer the case)
If you are in the midst of planning where to locate a data center though, the original article is well worth reading as it includes a number of useful links plus a further explanation (if one is needed) as to why you do not want to locate a data center in these places.
Posted by Wilbert Ingels on Thu, Feb 17, 2011 @ 01:00 PM
It takes all sorts to keep a data center manager happy. Here are my favorite ten things to put a smile on the face of your average server wrangler:
-
Efficient code. Inefficient software needs more processing power to run. It consumes lots of electricity and costs more money. As a result, budget-conscious data center managers love lean, mean code.
-
Noise-cancelling headphones. Data centers can be loud. All that ventilation equipment makes quite a racket. Noise-cancelling headphones provide respite from the whooshes and hums of servers at work.
-
Magnetic tape. It’s tried, tested and light on power consumption. Memory sticks and iPods may have taken over in everyday life, but tapes are still an important weapon in the data center manager’s armoury.
-
Blinking lights. To most of us, the lights on computer hardware barely register. But to data center managers, they signal healthy activity. Servers are working, data is flowing – and they can relax. Equally, a UID light can help them track down a problem quickly.
-
Hardworking servers. Some data centers are hiding armies of ‘zombie’ servers, wasting power while sitting idle. That’s why data center managers prefer to have just a few servers running near capacity.
-
Epic uptime. Every data center manager has a server that refuses to die. With an uptime measured in years, it’s seen staff come and go without ever being switched off. Whatever you do, don’t mess with it.
-
Cold weather. Keeping servers cool without inflating the air-con bill is a challenge for data center managers. As natural cooling works wonders, don’t be surprised if they have an eye on cooler climes. Though after recent seismic activity, perhaps not Iceland.
-
Locating a hot spot. Heat damages equipment and shortens its lifespan, so hard-to-find hot spots are a menace. Underneath server racks, behind cabinets – tracking down and eliminating them is a real victory.
-
Fixing problems quickly. Data center managers know that small problems grow. When something as simple as a broken air cooler can cause server failure, they need to move fast, using tools like our Smart PDU or "EnergySwitch" to resolve issues.
-
Good coffee. If electricity is fuel for your servers, caffeine is fuel for the people who keep them running. A caffeinated manager is a happy manager – just be sure to keep lattes strictly away from computer equipment.
Posted by Mark Harris on Wed, Feb 16, 2011 @ 01:00 PM
I have to laugh everytime I see a new offering in the 'Smart PDU' market. Somehow over the past 5 years, most of the manufacturers of rack based power distribution units added a Linux controller core and some number of power sensors to their premium PDU designs and decided to call this "Smart". Humm... Capable? Yes. Instrumented? Usually. But Smart?
No. Not really. Smart implies a level of intelligence and ability to take action, provide information otherwise not apparent. It's about Interpretation. Weather predication is smart. How many pop-tarts are needed on the store shelf at any given time to maximize profit is smart. Basically the concept of 'smart' takes advantage of lots of data points, applies some set of rules, and then provides interpretation results. That is smart. Is my car smart? Nope. Not really. It basically does what I tell it to do. Sure it's a complex system with lots of instrumentation, even a few controllers to provide operations and feedback.. But smart? Not really.
So why is a PDU Smart just because it has a controller and some instrumentation? It's not. It's the means to the end. It responds to any request. Ask it a question, it gives you the answer. Each time every time. YOU are smart, not the PDU. A simple electrician's analog multi-meter and a few minutes and you can get the same Voltage and Current reading that a so-called 'Smart PDU' will give to you. The current crop of Smart PDU products just allows this to be done remotely.
Now, in 2011 we are starting to see a new or next generation of Smart PDU, strike that, let's call them "Genius PDUs" which actually DO the gathering and interpretation of lots of data and present intelligent results. Their controllers don't just measure a metric (such as Voltage or Current) and report this value back as a RAW number, but they use high speed DSP technologies to sample thousands and thousands of energy points, run highly specialized Fast-Fourier Transformations (read: really smart) and provide actionable metrics to help data center manage their world. Imagine getting Voltage, Current, True power, KVAh, Power factor, hertz as well as all of the associated Quality metrics including ripple, spikes and sag. Imagine having these metrics gathered in a single 'cloud' system and being presented/actioned across an entire Enterprise. An Energy Cloud is really Smart, strike that, Genius! Just ask us to show you...
Posted by Wilbert Ingels on Wed, Feb 16, 2011 @ 08:00 AM
A recent issue of PROCESSOR Magazine had an interesting article (“Extend the Life of Your IT Assets: Strategies to Keep Assets Running Strong”) by Elizabeth Millard that is well worth reading because it outlined strategies that will help you get the most out of your IT assets. While so many people have focused on the logical layer for years, (those servers and switches and power-hungry IT devices), it is the physical layer that is worth a new consideration.
Overall key points from the article included the following:
-
“Focus on power and cooling efforts, because improper cooling can decrease the life of IT assets.”
-
“Look at SLAs to determine whether they’re meeting the needs of the data center.”
-
“Build in extra time for monitoring and planning to make lifecycle management a greater part of data center operations.”
Specifically though, one key tip was the need for integrated coordination among minor data center items (such as air distribution, cable pathways, equipment placement, fire
suppression, and lighting). However, Millard also pointed out that this is often overlooked – leading to a poorly designed data center that inadvertently puts more strain on some of your IT assets that are not being cooled properly.
Keeping that tip and the other key points in mind will go a long way towards extending the life of your IT and data center assets.
Posted by Wilbert Ingels on Tue, Feb 15, 2011 @ 01:00 PM
Clearly the industry as a whole is deeply concerned about using energy for data centers effectively. One of the most basic actions which is available to all data center operators, regardless of technologies and vendors installed is a renewed look at the entire power chain. How else would one know where a data center design is inefficienct unless there is a fundemental understanding of how and where power is being used? Modern Intelligent Power Distribution units gives data center managers this now required insight into power usage and more control over consumption. Using less power in the first place is a smart way to be greener.
But let's start at the beginning. The first thought worth noting was the point that many so-called “green” solutions for data centers are actually an oxymoron because they still have a carbon footprint and will also typically use power generated by fossil fuels. With that beign said, designing “greener” data center infrastructure technologies into any proposed new builds, renovations or upgrades for existing data center facilities is still critical.
The various blog posts have noted two important steps towards achieving a green data center:
- The collection of accurate measurements and environmental trends within a data center’s facilities in order to make changes and fine tune the efficiencies.
- The elimination of air mixing in data centers in order to reduce the power consumed by the HVAC systems that are supporting the data center.
In addition, the posts note that effective data center capacity management will be the key to any "greening" initiative. This capacity management will lead to:
- “Better predictability of space, power, and cooling capacity and redundancy limits means more time to plan on ways to mitigate their affect.”
- “Increased real-time availability of IT operations as a result of an enhanced understanding of the present state of the power and cooling infrastructure and environment.”
- “Reduced operating cost from energy usage effectiveness and efficiency as well as operator effectiveness from the use of automated tool sets.”
However, the posts end by noting that most facility oriented "green" solutions provide only a fraction of the efficiency gains that will be found in IT focused solutions such as server consolidation, virtualization and data reduplications. However, the post was also quick to note that facility orientated “green” solutions should still be considered if they fall in line with a data center’s design criteria and ROI needs.
Hence we all need to take a long hard look at how data centers are being designed, how power is being measured and how we plan to acheive responsible 'green' status. Essentially how we can benchmark our efforts to use power more efficiently. It all starts at the device, and is requires fundemental new approaches to monitoring and control of power and environmentals.
Talk to Racktivity. While there are many choice of 'Smart PDUs', we consider our PDUs to be GENIUS PDUs! Ask us to show you why...
Posted by Mark Harris on Tue, Feb 15, 2011 @ 08:00 AM
(Based upon a blog originally posted by Wilbert Ingles, CTO @ Racktivity)
When it comes to electricity use, datacenters have a lot to account for. Nearly every older data center today is being stressed past it's capacity in terms of power, and the natural reaction by many of these data center planners is to build new centers. Highly publicized projects are headlines almost daily in the IT journals.
Curiously, there is a better way to solve the power capacity problem and which directly helps DELAY these huge capital expenditures. Energy Efficiency awareness.
It turns out that these same companies could consider HOW they are using power, and have dramatic impacts on their overall capacity in EXISTING data center space.
The efficiency puzzle
There are lots of ways to make datacenters more efficient. Some aren’t practical – for instance, you can’t easily relocate an existing data centre to a cool climate. However, smaller measures can bring impressive improvements:
Shut down unused servers. It’s amazing how many ‘zombie’ machines are hiding in an average data centre.
Replace inefficient kit. That ancient server that sounds like Concorde taking off? Ditch it for something more energy conscious.
Replace any faulty power supply units in your servers. Before they fail completely, they often use lots more power than they should.
Smarter is greener
There’s a lot of wasted processing power in datacenters too. Plenty of them have rows of servers switched on, each using only a fraction of its total processing power. What a waste.
Intelligent software and hardware can fix this, turning off or throttling back under-used servers. At quiet periods, the datacenter can function with fewer servers online. Those that are switched on simply run at a higher capacity.
It takes clever hardware (like Racktivity's EnergySwitch PDUs) and software to seamlessly shift processing power around like this. But get it right and do the other things you can cut overall power consumption by up to half. That’s more money for your IT budget and less money wasted polluting the planet.