Microsoft and Google - Are they really ceasing data center expansion?

With the numerous reports of Google, Microsoft, and others halting data center builds you would think the data center business was caving in on itself like a dying star.

Town governments as well as the state Governments where the builds are halted are groaning. Sure they gave up tax breaks (their revenue) and the payroll tax that was supposed to come won’t, but I have to wonder why they are they that concerned. Data centers have a lot of servers in them, not a lot of payroll taxes in them, so at the end of the day they are out what they would have been out anyway which is the revenue on the construction and sales tax.

So the governments are feeling what the companies are – a contraction in the economy, driven by lack of access to capital in the credit markets. So things contract.

I can’t help but wonder though, with green technology evolving quickly AND the emergence of containers to house servers and their inherent improvement in efficiency, are they really thinking and retooling for the long term?

Both are smart companies, and I learned a long time ago that smart companies will take a strategic look inside their four walls when things slow down to retool and improve for when things do expand again.

I was in a meeting today discussing PUE of containers with an organization considering a build of a new data center. If you can make a data center TWICE as efficient, for 1/5 the cost, and shift the capex to almost 100% opex - wouldn’t you do it?

I think Microsoft and Google are simply rethinking how they want to deploy computing horsepower. Containers are a very modular, very fast and very economical way to deploy horsepower. I realize fully that those of us in the traditional data center realm are looking at containers and scratching our heads, however I for one believe that they are the future of data centers – which at the end of the day house servers with CPU and memory and give us resource pools or clouds.

If that computing resource pool can be deployed at massive density, double efficiency, and at a fraction of the cost – who in their right mind wants to defend the position of paying twice as much for something half as good?

Me neither…

I am working on publishing a white paper that explores the cost and efficiency of containers vs traditional data centers and will post the link when it's done in a few weeks. If you care to share data - please do. I will change the names of the innocent if needed - I just want to make sure I put out real data.

Sun's new data center

I just read a story about Sun Microsystems' new data center in Broomfield, CO. There were some interesting highlights in the blurb.

Reduced Electrical Consumption: By 1 million kWh per month, enough to power 1,000 homes in Colorado

While this is a great stat, where did the savings come from? What is the PUE? Sun rolled out the Blackbox product last year - Containers/Pods are the next wave of computing as far as I am concerned - and I'll bet the container is still more efficient and the power saved in a Blackbox would power 1,000 homes on the cooling alone. You double the cost and draw of electricity in cooling a data center so unless the environment is optimized to support heat dissapation and optimize cold air flow, you don't gain much.

Reduced Raised Floor Datacenter Space: From 165,000 square feet to less than 700 square feet of raised floor datacenter space, representing a $4M cost avoidance.

I used to walk through a raised floor data center every week in a previous job and I always wondered why they were all the rage. The amount of cold air under the floor that didn't do anything was astounding. I know the electricians had a better environment to work in, but at the end of the day, no one ever asked me 'What am I paying for my share of idle cold air?'. Cooling is a necessary evil, and an expensive one. Looking not at cooling but how heat is removed is the key.

Enhanced Scalability: Incorporated 7 MW of capacity that scales up to 40% higher without major construction

What the **** does that mean? Are they using high octane diesel in their Caterpillars? Funny car fuel? A four barrel carbuerator?

Superior Cooling: The world’s first and largest installation of Liebert advanced XD™ cooling system with dynamic cooling controls capable of supporting rack loads up to 30kW and a chiller system 24% more efficient than ASHRAE standards.

I love Lieberts - they work, they're reliable, and they mask poor design. In this series of Lieberts it appears that their offering is more modular, and more deployable into trouble/hot spots than its other offerings. That is great, however they do not put the draw into the technical documentation. Efficiency is great, but if it takes 30% more power to generate 10% cooler air in the same footprint, I don't see the value.

Greener Architecture: Including flywheel uninterruptible power supply (UPS) that eliminates all lead and chemicals waste by removing the need for batteries, and non-chemical water treatment system, saving water and reducing chemical pollution.

I wonder when someone will actually have the stones to say, Green is great until it doesn't work or it costs 1.5x what traditional (proven) technology can deliver. Case in point: I wonder what the carbon output of manufacturing a wind turbine is compared to the .3 megawatt it generates. The steel manufacturing (they don't use woodstoves or solar to melt metal), the wiring manufaturing (people dressed in hemp clothing, using bamboo shovels don't find wiring deposits), the lubricants, the shipping fuel... Blindly going green because it's trendy may not be the answer. Yet. Did you cut down to one square of toilet paper to save a tree? Me neither.

Overall Excellence: Recognized with two Ace awards for Project of the Year from the Associated Contractors of Colorado, presented for excellence in design, execution, complexity and environmental application.

But not efficency? How can give an award for data center design without efficiency being the most weighted category for judging it? It is like giving out an award for best dessert without tasting it.

I love that companies like Sun are pushing the envelope and I hope it continues. I just hope that common sense enters the equation and that an actual yardstick is used to measure what matters.

A Penny Per GB of Storage

One of the comments I got back from Verari was that their solution is capable of 13 Petabytes of storage, going well beyond the 600 TB I used as a benchmark.

So since a single Petabyte is 1024 TB, you basically double the storage for the sake of math and keeping it in line with my earlier benchmark of 600TB:

Petabyte = 1024 Terabytes

The new computations then look like this:

At ~$200 KW, you’re looking at $120,000 month in base rent for a 600kw solution ($200*600)

So for 13.3M gigabytes you’ll pay about 130,000 per month (keeping numbers round) which means you’re at one cent per gig on cost. And more efficient and that is with the drives fired up 100% of the time.

Microsoft needs to implement these to archive email at these prices... Or Live Apps data.

Mark

Low cost 600TB backup...

I forgot to include this little tidbit too in my previous post...

If the bandwidth charges were whirring through your head like the electric meter on Clark Griswold's house after putting up the Christmas lights - try this:

When the container is ordered ship it to your office loading dock. Plug it in, and throw a strand of fiber out the window and plug that in. Voila, 600TB at the speed of light, no bandwidth charges. It's a bit of a hack, but, cha ching!

600 Trillion Bytes per box - ROI of Pod Container Storage

I read recently that the IRS stores 15 to 20 Terabytes of data for each year of tax returns.

If the number is 20 Terabytes, 600/20 = 30 years of tax records storage is possible with a single container/POD. Since they go back 10 years for most practical applications of retrieval this means, at least conceptually, that the IRS and any other entity that stores records electronically (lawyers) will pay for 66% of records they seldom or never need to access for legal, operational, or other reasons.

For other organizations (banks, brokerages) it is 4-7 years where records need to be accessed on a 'regular' basis and after four years the trend falls off a cliff and after 7 it's hello abyss. So let's look at what a company might pay to store all this data and then let's look at what they would save if they thought differently about how they actually implemented their storage both in dollars and in energy efficiency.

1024 Gigabytes (GB) = 1 Terabyte (TB). I will round it to 1000 to keep it simple.
I will also assume that the cost per GB is $.50 - 50 cents US, so cost per TB is $500 USD. So that means in raw Capex you're at $300,000 ($500*600) for a container of raw storage.

Let's look at Opex now. We have power, cooling (subset of power), and space.

Power for a container is 350-600kw. If I use 600 or the max draw/densities I have seen, then 1TB = 1 KW of power for keeping the equipment fired up. At $200/kw or $200/Tb you're at $120,000 per month in base rent. Here is where the containers leave the data centers in the dust - PUE.

A data center PUE (Power Utilization Efficiency) looks at how much additional power is needed to power the cooling and facility in which the computing equipment is stored. Most data centers are at 1.7 for a PUE, meaning for every Kw of power needed to power the computers, an additional .7 Kw is needed for cooling it and ultimately powering the chillers, the A/C, etc.

POD Containers are 1.08 to 1.3. What does this really mean?

That you will save .4 to .6 per Kw in a container per month in power alone.

In real dollars that means per TB of storage, saving .5 Kw (mean average) equates to cutting my Opex by about a 1/3 over a traditional data center FOR THE SAME FUNCTION - storage of data.

So we lop off $40K/ month, which means that over a year we save $480K ($500k to round it). This covers the cost of the storage ($300k) and then some. You can look at it as free storage plus $200K, or a substantial reduction in Opex, or $200K to do other things with.

Bottom line is a POD/container pays for itself in under a year, is FAR more efficient than a data center day to day and that's not only good for our bottom lines, but it's good for our environment, and is responsible computing.

I will take a look at an even smarter solution for older records using MAID - Massive Arrays of Idle Disks.

Want help crunching your numbers? Email me -

mark.macauley@gmail.com

How much is One Trillion

With all of the talk about a Trillion dollar budget/bailout/deficit/cost of Iraq I had to try to wrap my brain around it and here is what I found for explanations:

A trillion dollars is so large a number that only politicians
can use the term in conversation...

Here is some perspective on TRILLION:

1 TeraByte (TB)

Trillion = 1,000,000,000,000.
The USA has not existed for a trillion seconds.
Western civilization has not been around a trillion seconds.
One trillion seconds ago – 31,688 years – Neanderthals stalked the plains of Europe.

or this:

Let's start with one billion. If you spent $1,000 per day, every day without fail, you would spend $365,000 in a normal year.

You would spend $36,525,000 in a century (assuming 25 leap years per century).

So, if you started spending $1,000 per day starting January 1, 0000, you would be spending $1,000 per day through the end of the Roman Empire, the collapse of the Mayan civilization, the Middle Ages, the Crusades, the Black Plague, the Renaissance, the European conquest of the New World, the Industrial Revolution, the American Revolution, the French Revolution, the Russian Revolution, World Wars I and II, the Cold War, the end of the 20th Century and would finally reach one billion dollars a little more than one-third of the way through the 28th century.

Now do that 999 more times and that gets you to one trillion dollars