Mark Bramfitt - Utilities & IT

Facebook Has Moved From MAID, to well...MAID

In my courses on data center energy efficiency, we touch, too briefly, upon data storage technologies that can reduce energy use. I say too briefly because data storage growth rates vastly exceed compute growth, and in my mind not enough attention is paid to the half dozen or so energy efficiency measures and best practices.

One of my more enjoyable anecdotes was to talk about MAID storage. The technology was pioneered by Copan about five years ago, and despite a similar offering from Hitachi, it hasn't seemed to gain any traction in the market.

MAID systems are really all about controlling disk drives - a software front end determines whether the data to be stored will be accessed often or rarely.  The rarely-accessed data is stored to disk drives that are then shut down - you can get the data again, though it takes a little longer to get to it as the drive has to be spun back up.

A brilliant system and very easy to see the efficiency gain: in Copan's system no more than a quarter of the drives could be on at any time, so energy savings was 75% compared to normal cabinet full of drives.

The anecdote was that Facebook was an early adopter of the technology, so that meant that the profiles of unpopular people were being sent to the idle disks, while popular people had profiles on live disks. Funny! Of course in reality Facebook was probably simply saving older posts to the idle disks, as people don't really search back very far in profiles.

Now comes word through DatacenterDynamics Focus (page 64 of the current issue, subscribe here) that Facebook has brought the MAID concept in house, though in a new way. Facebook now sends little-accessed data to "cold storage" - disks that are turned off, but also housed in a separate data center. 

The use of a separate data center makes a lot of sense when you think about it. There is no need for backup power for the cold storage racks; no UPS, no generators. The racks likely need far less cooling capacity than "standard" storage or compute racks too.

Will this approach work for enterprise data center operators? I don't think so - it's a pretty customized solution that will work for large-scale operators like managed services and cloud computing sites, but I doubt many enterprise users will be able to justify dedicated data storage sites. And again, it doesn't look like the MAID solution has any traction at all.

I expect that solid state storage solutions will be the primary go-to for enterprise applications, along with data management measures (deduplication, compression, thin-provisioning, etc.).

Now let me get back to reworking that anecdote...

 

HP Looks to ARM for a Leg Up on Server Efficiency

HP announced today that they will be offering servers in the middle of next year that use ARM chips (and eventually, Intel's Atom processor). The stated benefit is primarily energy efficiency, and HP figures that maybe fifty of the largest data center operators (those with web serving applications) will be interested in the product.

I dug around the web to find out as much as I could on the "Moonshot" line; this coverage in the New York Times sums it up pretty well, HP's press release is here.

Note the description of the energy use of HP's system compared to a "base" load of 400 servers - energy use drops by almost a factor of ten, from a load of over 90 kW to just over 9 kW. The space required goes from ten racks to just half a rack, so the new architecture pushes the density envelop into the 20 kW per rack range, which can be cooled with air, but is likely to push operators to consider at least close-coupled cooling solutions.

It will be interesting to see if the product line makes headway, and whether other users may adopt it as well. HP notes that scientific and financial processing loads, which require intensive data analysis, might be suitable for the equipment also.

And it appears that we have two fundamentally different approaches to server energy efficiency coming from HP and Dell, with Dell offering more traditional architectures but warranting equipment to much wider environmental conditions - essentially offering equipment that can allow air-side economizer use all of the time, anywhere in the country. HP is countering with fundamentally more efficient equipment, but with the potential liability of high energy density.

 

Samsung/Dell: Dell Wows with Higher Inlet Temperature Certification

Perhaps the most earth-shaking news at the Samsung/Dell CIO Forum was the announcement by Dell that they were certifying their data center IT equipment to much higher inlet air temperatures - sending the signal to data center operators that the equipment is warrantied at much hotter environmental conditions.

In the data center energy efficiency programs that I have delivered over the past three years or so, we have always used Dell's specification sheets to show attendees that manufacturers warranty their equipment at temperatures well above the ASHRAE guidelines.

We cite a specification for a Dell Powervault MD3000, which lists suitable inlet temperatures of 10 to 35 degrees Celsius (50 to 95 degrees F).

Now Dell has announced that their servers will be warrantied in environments that reach 45 degrees C inlet for no more than 90 hours a year, and that their networking and storage gear can have 40 degree C inlet air for as many as 900 hours a year.

This is big news, because it removes any possible obection to using nothing but outside air to cool a data center in any climate zone in the United States, and I suspect the specifications were set expressly for that reason.

Why? Dell showed that they are building modular data centers to support Microsoft's Bing search service that in an installation in Colorado went from bare ground to operational in four months. Dell claims a PUE of 1.03 for the 1 MW deployment, which can only be achievable using 100% outside air cooling. Dell also uses servers without onboard fans to get this amazingly low PUE rating.

Samsung/Dell: Samsung Touts SSD for EE and Performance

Some analysts that I've spoken with recently believe that we're at a technology adoption threshold where solid state memory will begin widely supplanting traditional disk storage devices.

Samsung certainly thinks so. At the CIO Forum held in conjunction with Dell three weeks ago, they reinforced their entree into the data center space, showing in particular how SSD can greatly reduce energy use in servers while dramaticaly increasing potential performance.

Samsung has of course made SSDs for mobile devices for a long time, and with their projection that 10 billion mobile devices will be sold globally over the next decade, that's clearly a driving market for them.  By the way, Intel projects that there will be over 50 billion internet connected devices in use in that timeframe.

With SSDs now made with 30 nm architecture, Samsung notes an inherent energy efficiency advantage approaching ninety percent compared to disk, with follow on savings from reduced cooling load in a data center environment. As to performance, the input/output rate for SSD is an order of magnitude better than disk, theoretically allowing servers to use their full processing power.

(I'm inclined to think that the later advantage is valuable only for certain users and applications, given that most servers run at ten or fifteen percent utilization in enterprise environments.)

Samsung's presentation at the event was really solid, with a great overview of the energy use issue facing data center operators. This product brochure concisely tells the energy efficiency story on the second and third pages.

 

Green Grid Weighs in On Utility Virtualization Incentive Programs

I had the good fortune of attending a portion of the Green Grid Technical Forum in Santa Clara last week, and seeing Corban Lester's presentation based on a new white paper that analyzes utility server virtualization/consolidation incentive programs. (Presentation only available for Green Grid members.) Corban is with Lockheed Martin and heads up the Green Grid's Utility Task Force.

The paper and presentation did a reasonable job explaining why utilities have entered the data center and information technology markets with their energy efficiency programs. I have quibbles on minor points: I don't think efficiency program managers are as risk averse as portrayed, or that utilities are attracted to the market because it has a high PR profile. Utilities are scrambling for savings to meet their targets, and have an obligation in many cases to address all energy use sectors.

What I do support in the paper is the conclusion that utility incentives for server virtualization and consolidation is still appropriate despite decisions by California utilities (and BC Hydro) to end their programs.

A tailored program offering that targets smaller enterprise customers rather than sophisticated IT shops is still needed to drive market penetration and adoption of virtualization - my reading of the market reports says eighty or ninety percent of the market is untouched. That represents an incredible opportunity to drive energy efficiency in the short term, slowing the need for new data center capacity.

New Power Conversion Co Funded by Google Targeting DCs

Interesting squib in the New York Times Bits blog late Wednesday regarding a start up company in Goleta called Transphorm, backed by Google and other venture capital firms.

The company promises to deliver power conversion "modules" (that sound a lot like transformers with AC/DC rectification - like the power bricks for laptop computers) based on gallium nitride instead of silicon. (And despite double majoring in Materials Science and Engineering, don't ask me about the science!)

Of interest is Google's backing, with the stated intent of applying the technology to power conversion equipment in data centers, although Google has yet to test any prototypes.

The energy savings potential noted by Transphorm is huge - hundreds of terrawatts, though that appears to be for the entire energy conversion market, certainly not for data centers alone.

With "standard" power supplies in servers and IT equipment in the mid-ninety percent efficiency range, this technology is certainly chasing a small sliver of remaining potential, but then the market is immense. It makes sense to go after data centers as an entry point due to load factors that would yield high rates of return. 

It will be interesting to see the technology in action, and if it can be cost effective compared to traditional high efficiency power supplies.

Thanks to Agrion Panel & Recap

I was asked to moderate a panel yesterday in Palo Alto hosted by Agrion, an organization that hosts discussions and events related to sustainable development and cleantech (registration and membership required to access content, including recording of the session).

Our topic was data centers and energy efficiency, and the dynamite panel included representatives from eBay, Extreme Networks, NetApp, Oracle, and the Wired Real Estate Group.

We managed to cover a lot of ground in an hour-and-a-half; as you may know from attending events like this each topic naturally raises several promising lines of discussion, so it is hard to cover all of the allotted topics.

When it came to "matters of religion" we had some good debate about the use of air-side economizers in data centers (consensus view: they should be mandated in all new facilities but are hard to retrofit into existing ones), the move to high or extreme rack densities (it's going to happen, so designers and operators should plan for it), and data center efficiency metrics (PUE is good enough to evaluate the facility, and no one knows how we can get to an IT efficiency index).

In conversations both before and after the panel time, I talked to folks who recognized that industry leaders (such as our panelists and their companies) are way ahead of the curve compared to typical enterprises and other geographical regions of the US. It's almost becoming silly to talk about data center efficiency in Silicon Valley; we have to continue to take the message out on the road.

Finally, I talked to am environmental quality rep from a major IT company who said it is about time to talk about the use of IT to drive improvements in other areas of the economy. In other words, maybe data center and IT growth, certainly if implemented with energy efficiency as a priority, is on balance a great thing, if it helps us manage energy use better in other areas of the economy.

My personal thanks to Dean Nelson, Kevin Ryan, John Wallerich, Brian Day, and Franz Seidelhuber for a great discussion.

The Jevons Paradox and IT Energy Efficiency

The New Yorker published a lovely piece by David Owen under their "Annals of Environmentalism" banner in the December 10, 2010 edition. (Subscription required for access to full article; abstract free.)

The article describes the Jevons Paradox, which postulates that improving the efficiency and thereby lowering the cost of a product or service will result in more of its use rather than less. The cited case in the original paper by Englishman William Stanley Jevons were efforts to improve the efficient use of coal.

Another English economist (Len Brookes) brought the concept back to the forefront during the oil crisis in the 1970s, arguing that devising ways to do more with less oil would result in higher energy consumption, not lower - a so-called "rebound effect".

Discussing this paradox and determining if it is valid is of critical import to energy policy and efforts to improve energy efficiency, so is of obvious interest to utility professionals in the EE field. I expect that IT professionals will be equally intrigued, given the tremendous rate of change in the performance and yes, efficiency, of successive generations of IT equipment.

It is not an unfair question to ask whether improving data center or IT equipment efficiency will simply make all IT capabilities cheaper, and therefore used more, and by more people, gobbling up any energy savings and attendant environmental benefits.

I've got the outlines of a paper (or perhaps a chapter in a book I am pondering writing) that specifically looks at IT and energy use and the Jevons Paradox.

In the meantime, if you want to dig deeper, I recommend this post from James Barret at the National Geographic website, and not only because he reached for the same Yogi Berra-ism that I did when first thinking about Jevons.

 

Calling Out HP for Poor Marketing Ethics

Only peripherally related to IT, so disregard if you wish...

HP has fallen into the tender trap of allowing their marketing agents to run roughshod over civil conduct and indeed civil laws in San Francisco, by posting flyers on every single lamp pole on Lombard Street in San Francisco over night.

Yes, I have a personal pet peeve over visual litter and the co-option of public space for commercial use.

Funny, but IBM got caught out eleven years ago when their ad agency stenciled advertisements all over Financial District sidewalks, eventually paying more than $100,000 in fines and restitution.

What should HP do (in the unlikely event they're listening)? Remove the offending flyers, by tomorrow morning. Publicly apologize. Flay their marketing firm or better yet ban them from HP work for evermore.

Hard to believe that marketers think these activities do anything but engender customer hostility, and harder still to believe they can foist plans like this on otherwise responsible companies like HP.

 

 

Recommended Reading; Yes, In Esquire!

Laugh if you must, but I find the editorial content of Esquire (and GQ) superb, and am recommending an article from the December issue related to data centers and computing. Author Douglas Fox introduces us to Stanford professor Kwabena Boahen, who is studying neuromorphic engineering.

Check out the energy use estimates in the sidebar, comparing the human brain to a desktop, and to data centers. Note that even if we had the code to mimic human brain function and capacity, we'd need a data center drawing 60 MW to handle the computations and data storage.

Fox does a damned decent job explaining the energy issues of data centers (about two thirds of the way in). Not to be skeptical, but the claim that an internet search uses the equivalent energy of burning a match harkens back to the discussion in the industry a year or two ago where people were saying it was the equivalent of heating the water for a cup of coffee.

I'll leave all of that to the experts, but I do agree with the premise that there are "epic challenges ahead for the computer industry" related to energy, and that a likely combination of energy efficient technology and perhaps whole new ways of performing computational tasks will be needed to make the next leap forward.

Addendum: I'm embarrassed to say that the Esquire piece is essentially a partial reprint of an article that first appeared in Discover, and has also been reprinted in The Best Technology Writing 2010. Both are a bit longer, and include material on IBM's interest in neuromorphic computing.  Kudos to Douglas Fox for leveraging his work through multiple outlets.