Google Efficient Data Centers Summit - 3 of 3

Uploaded by Google on 07.04.2009

>> HOELZLE: A brief overview of Sustainable Operations by Bill Weihl, our green energy
czar, and then followed by a discussion of water usage in data centers and other very
cool video of an actual Google Data Center Internals. So, Bill, please take it away.
>> WEIHL: Thanks, Urs. Okay. So, it's actually seems quickly appropriate given the discussion
at the very end of the last panel and seeing the comments that Ken was making. One of the
questions would be, what is sustainability and does it really matter? Should we be thinking
about it? And I think today is all about energy efficient data center. So the focus is really
been on energy, we'd certainly heard about water and some other issues. Hiding behind
energy, implicit in that, is the climate impact of producing all that energy. I believe and
I think Google believes that people at Google--most people at Google believes that we have a responsibility
to worry about sustainability, to worry about the climate. And that, I think Olivier was
absolutely right that if you think about this, you might initially think, "Well, it's going
too cost too much so I can't do that." But if you think about it, you can sometimes come
up with creative ideas that turn out not to cost too much, or the things that help you
with sustainability turn out actually to save you money, in which case, they would have
been a good thing anyway. So I think it's important to focus on this issue, maybe not
have everybody in the organization focus on being--you should be thinking about it. So,
what is sustainability, though? So, I did a quick Google search. I don't think I'm allowed
to say I Googled it, but the lawyers will tackle me or something. What is sustainability
mean? And I'm not going to read this, but I think there are a couple of points to take
from this. One is it's about not depleting resources, so preserving the planet's ability
to support us and our civilization and ecosystems into the future. And then the other which
really (ph) relates to business is that sustainable businesses are about what people called the
triple bottom line, which is a phrase, to be honest, I kind of hate. But it's about
the point that if you want to have a sustainable business, you've got to worry about economic
sustainability and growth and prosperity for the business. You got to worry about social
aspects for the community that you're in, and you've got to worry about the environment.
All of those are important. You forget any one of those and in the end it's probably
not actually all that sustainable for the long term, okay? So, I want to talk just briefly
about best practices, then I'm going to touch on a couple of issues, and then Joe is going
to dive deep, you know, into the deep ocean, more or less on the water side. So, we've
heard a lot of this today already, so I'm not going to dwell on it too much. Clearly,
measuring is important. If you want to be more sustainable, you got to figure out what
to measure and you got to measure it. If you don't measure it, you can't manage it. Given
measurements, you can prioritize and, in fact, to decide what to measure, you need to a first-level
prioritization because some things are not significant enough to be worth the time to
measure or to measure very accurately. So there's prioritization that's going to go
on all along the way. And we've been doing this; I've been at Google now for three years.
We've been gradually focusing on more and more aspects of our operations and how to
make them more sustainable. You really need to think about all the resources, so we talked
a lot today about energy and we've talked about water. But you got to worry about wood
and metals and any other materials that are used to build or produce your products and
services. And you've got to look at the entire supply chain upstream, your suppliers, also
downstream, your users, a full life cycle to understand what are the environmental impacts,
what are the social and economic impacts et cetera, et cetera, okay? "Embodied energy"
is a phrase that people use a lot of it, the energy basically used to make something. So
you buy a laptop, the embodied energy in a laptop is more than the energy that will be
used during its operational use phase in its lifetime. Just kind of striking and that's
because it's a very energy-efficient device. It's probably true for a cell phone or even
more so, right, that takes a lot of energy to make the stuffs that goes in it. So that's
something and once you've made the thing itself energy-efficient, you better start thinking,
well, what about all the energy that went into producing it? You need to aim for continuous
improvement. I think it was Olivier who said this just a minute ago. It's not about the
destination. Perfection is not the goal. Improvement is the goal. Getting to a lot better than
where we are today is the goal. And then, one of the things, and this gets to sort of
the social aspect of the triple bottom line, one of the things we focus on a lot here is
keeping our employees healthy and happy. We try to build workplaces that have good indoor
air quality, good lighting, in other ways are places that are comfortable and healthy
places to work and there are fun places that people are happy that helps them be, you know,
healthier et cetera, et cetera. And that's an important thing. I'm not going to focus
on that though, you know, if you saw the cafes, you may have seen, you know, I don't know
do we have any of the big bouncy balls in here, you know, massages et cetera, all that
stuff is good for making sure that, you know, we stay healthy and happy working here. So,
I'm going to touch on two issues. First one is E-waste. It's a growing problem and I don't
need to go into why it's a problem. We produced a lot of stuff in the industry and, historically,
over the last decade or two, a lot of it is going to landfills and a lot of it is going
to the landfills in developing countries; this is a disaster. What are the solutions
to the problem? That's what I would call the four R's. Reduce, so they don't just buy new
stuff because it's new and you can, buy because you need it. Reuse what you got and either
yourself or get others to reuse it. Repair it if it's broken and then reuse it, and then
recycle whatever is left, don't just landfill it. And whatever--you know, sometimes there's
going to be some amount of stuff that you can't do any about with. Well then, that you
got to find some responsible way to dispose it. So what do we do in here at Google? When
our servers reach the end of their useful life, for us--well, first of all we reuse
68% of the material and retired servers. Memory chips, CPUs, disks and other stuff gets reused
either by us or by others. The remainder is recycled into raw materials, essentially,
by smelting it to extract all of the useful materials and precious metals, et cetera,
and to, you know, get rid of the rest in the form that can be safely and responsibly disposed
off. We don't throw stuff in open landfills, we don't just incinerate it and have toxics
go up into the atmosphere, and we don't export it to developing countries. And this is true,
so if you look at what to reuse or to recycled, that's basically a 100% of what we get rid
of. And we don't just store stuff in the warehouse waiting for some future solution. If you look
at what's happening in the U.S., the EPA estimated in 2006, only 18% was reused to recycle, the
rest was in land filled incinerated or stored, waiting for some future solutions. So solutions
are here today for this problem and lots of other companies actually in the industry that
could make huge strides in the last two, three, four years on actually getting more toward
this point. But there's still an awful a lot of people who aren't taking advantage of those--people
and companies that buy equipment or individual consumers. They now have good ways to dispose
those stuffs responsibly. I think that's an issue that's been I think a growing problem,
but one or maybe we've crested it, the hump and their coming back down the other side
where, as an industry; things are definitely improving a lot. The other thing I want to
touch and this would be really quick because you've heard about it from James and a bunch
for other people is water. This actually is something that is a big issue in the Western
U.S. There's a huge drought coming. I just put this headline up here because it sort
of struck me that ESPN had an article recently about the coming U.S. crisis in water policy.
It never occurred to me ESPN would actually be talking about that, but I guess that if
you're a sport fisherman you'd understand that. But the Western U.S.--there's a disaster
coming, you know. And we're entering what may actually be a historically really bad
drought and the water allocations from the Colorado River and other watersheds were based
on a time of exceptionally high historical water levels. So, we're just--the supply and
demand don't match. It is a problem for anybody who uses water, Google or anybody else, who
use water for any operations, much less, individuals. So, in fact late May, there's a prediction,
it came out of the scripts about a year ago that the water and lake may could effectively
dry up in terms of what is used for power generation, irrigation et cetera by 2021;
50% chance of that happening. That's a disaster. So water is clearly something we got to worry
about. Climate change may make this all worse, and I think James is absolutely right, if
you can avoid using water completely that is, by far, the best solution or if you can't,
then minimize the amount and Joe's going to now talk about what we're doing on the water
front. So, Joe. >> KAVA: Thank you. So, good afternoon, everyone.
My name is Joe Kava, I'm the director of operations in our data center group here in Google, and
I want to thank Bill for introducing the sustainable lead topics to us. And I also want to thank
James and Olivier and Dean (ph) and Andrew this morning who made some of my points that
I'm going to be making in this presentation. So, I'm glad to see that the donation envelopes
that I gave to you, guys, this morning actually made it into your pockets. But I am a sport
fisherman and I do read ESPN and I watch the ESPN and I watch ESPN and it is important
to me. And I do want to spend a few minutes delving into sustainable water management
with respect to data center operations. Dean [PH], I'm going to argue with you a little
bit. I think the big elephant in the room is water and maybe not only application utilization.
think that, in fact, this is the next big issue that all of us are going to have to
address. I think as a whole, our industry has done a very good job over the last couple
of years specifically with leadership from the Green Grid and the Uptime Institute and
all of you in this room that we've been focusing on power and power consumption and energy
efficiency, and that's excellent. I think the next thing we need to turn our attention
to is what do we do about the looming water crisis and how are we going to address that
with respect to sustainability for our data center operations. So, a few key overview
messages, I really like the way that Olivier was discussing that it's not simply just about
ROI, but it's about doing the right thing environmentally, and I take it a step further.
I think it goes into the TCO model and that--what about business continuity and that is a very,
very big driving force in your TCO model. So, I think that business continuity factors
are really driving some of our considerations for water resource management. I don't think
that any of us would argue that water resources are becoming more scarce and, as such, I think
the link between data centers and water resources is going to be more plainly evident to a wider
audience just as the link between power consumption and data centers has become more widely obvious
to a broader audience, and with that, will come public scrutiny and political--perhaps
political intervention. Andrew alluded to it this morning; water consumptions in data
centers might be regulated. I want to talk and maybe present a little bit of a different
viewpoint here in that reducing water consumption in data centers is a great objective that
we should all be looking for. But I think there's another opportunity, and at the right
size and scale, I think that reuse water and reclaim water use in the datacenter is an
opportunity, and I'm going to get to a little bit more details about that about what we've
been doing in that regard. I think that all of us agree that we're probably in this room
because we care about the environment and we care about the impact and the climate change
and water--or water use and water consumption has a big impact on that. So I dint think
there's really any argument about the impact of depleted water resources and the effect
of our environment. There are some other risks that I think that I propose we need to take
into consideration when we're talking about sustainable water resource management. Those
risks are availability due to drought or due to regulatory action, urgent access. It could
very well be that your data center expansion, the gating item for that expansion is the
schedule of your local water authority for upgrading their capacity in their lines. And
also competition from neighboring developments, right? So, if there's another big company
or another big industry coming into the neighborhood and they're going to be competing for the
same water resources, you may actually wind up having to reserve and pay reservation fees
for access to that water for your data center expansion that you planned. And future costs,
really, this capacity upgrades especially as water resource have become more scarce
and more difficult to extract, the upgrades that the utilities are going to have to deal
with to get at those water resources are going to be very expensive, very costly, and we
all know that those costs will be burdened by the consumers. But is this situation worsening,
or is it just status quo? So, this was taken from the Sandy Labs [PH] report to congress,
and it's a reprint of a table that shows water resources at various different regions across
the United States. And in each of these cases, the ground water and the surface water declines
are really just staggering. But at the same time, the USGS is recording that the total
water consumption in the US in 2000 was approximately the same as it was in 1990, even though the
US population has grown almost 13% during that same period of time. And they're attributing
that mostly to better technology, advancements in technology, and conservationism. So, if
the total consumption in the US is staying the same, is the problem that is not really
getting any worse? Well, maybe. So, these are two side-by-side shots of Lake Lanier
in Georgia taken from March of 2007 on the left to February 2008 on the right. And you'll
see how much the water level has receded in just that eleven-month period. 2007 may have
been a particularly dry year, though. So maybe it was just a temporal effect, maybe not.
This is the--another lake in Georgia also, in 2008 just last November, and what you're
seeing here are boat docks. Unfortunately, these were not intended to be dry docks. These
were actually up water level, just the past couple of years previous. So, the situation
may, in fact, be getting worse.
Just last week, the Wall Street Journal published a very timely article in which they claimed
that the total water withdrawals in the United States, of the total withdrawals, nearly 50%
of that is used in the electric power generation industry. Now, most of those water withdrawals
used for electric power generation are returned to the waterways at some point. But, still
2 to 3% are lost through evaporation. It's about 1.6 trillion gallons of water a year
lost to evaporation or roughly the equivalent of about 17 1/2 million U.S. households worth
of water use annually. So, why do we care and how do data centers fit into this? I'm
going to argue a little bit with James on this one. You use the 15 megawatt example
and what the power or the water consumption was for cooling tower evaporation and below
down. I'm going to claim that your numbers are wrong. The reason I'm going to claim that
is because if you look at the embedded supply chain and how much water was consumed in generating
that 15 megawatts of power, that the numbers are far more--far more significant. I normalized
it to a 10 megawatt IT load or a 10 megawatt cooling tower load. The average data center
would use about 150,000 gallons of water a day to cool that load in evaporation and blow
down. Up stream of that data center, the same 10 megawatts consumes about another 480,000
gallons of water. So, the total embedded supply chain is over 600,000 gallons of water per
day for that 10 megawatt cooling tower load. An interesting fact that as we are going through
this analysis kind of struck us is that because of the overwhelming influence of the power
generation factor in that that consumption that the average water list data center would
actually consume about a third more water than the Google evaporative cool of data center
with, you know, all things being equal, same 10 megawatt normalize load and all that. And
of course, PUE is a driving factor in there. But it really is because of the power consumption
ratio. So, in fact, what we've been doing, reducing our PUE is one of the most significant
terms you can use to drive down your overall water consumption as well. Maybe not something
that it is intuitively obvious, but all the work that we've been doing on getting more
efficient with power consumption is helping with respect to water consumption as well.
So, what's the cost of all this water? I'm going to argue that it's meaningless, it's
trivial, and that right now water is cheap. And, you know, the regional prices range anywhere
from less than a dollar for per 1,000 gallons to more than $5 dollars per 1,000 gallons.
It just doesn't matter. It is cheap. It doesn't show up on the top 10 things on your expenses.
I think that James--when you talked about the OPEX breakdown of a typical, you know,
data center, water doesn't show up on there. So, your CFO might not be targeting water--you're
doing about water reduction because it's costing you a fortune. So, I argue that the cost of
water is trivial from a simple OPEX standpoint. The cost of water from a business continuity
standpoint, and the access to that water, I argue, is extremely, extremely expensive.
So, what is Google doing about this? Since our data center's program--our internal data
center's program began in 2005. We've constructed two sites that are a hundred percent self
sufficient for water cooling process load. One facility uses a side-stream plant to treat
the effluent from the public utility, and use that in our cooling towers. Then we re-treat
the water and release it back into the public utility. A second site in Belgium--and I'm
going to give us a video tour of this in just a moment here uses water from a public industrial
canal. We treat the water in an onsite treatment plant and we use that for our cooling tower
and our process load. We are also currently constructing additional sites with water retention
ponds so that we could reuse storm water. Recapture and reuse that storm water. And
we've been really, really focused on lowering our PUE, lowering our total energy consumption
as a driving component in lowering our impact on water use. So, our goal, our company goal
is that we want to achieve 80% self sufficiency for data center water use by the end of 2010.
So, I'd like to give us a short video tour of the water treatment facility in our Belgium
data center. Ops, there's supposed to be a sound here.
>> Sound here. >> KAVA: Yes, it is. Okay. At our data center
in Belgium, the Nimy-Blaton Canal on the north side of the site surges the source of the
raw make-up water. The intake chamber is protected by a steel dolphin unit which ensures the
intake will not be blocked or damaged by ships or large debris. The intake chamber has two
stainless steel intake pipes with isolation valves. Each intake is sized for the 100%
of the total site capacity. A compressed air tank periodically discharges across the bar
grates to dislodge any material that may accumulate. Downstream of this intake chamber is the metering
chamber to allow access for the local water authority. From here, water flows to the distribution
chamber where initial solid settlement and hydrocarbon sensing takes place and then flows
into the 2,000 square foot water treatment plant. The plant consists of three main lines,
each capable of treating cooling tower make-up water to one-third of the critical capacity.
A combination sand, oil and grease separator receives the water as it enters the plant
from the sub-grate piping. Irraters are used to disturb the water during the separation
process. A conveyor removes the sand from the fluid and the suspended solids are skimmed
off and deposited into containers to be recycled. With most of the solids removed, the water
is passed over a Lamellic clarifier so that remaining particles will settle out. This
clarified water then flows over V-notch weir and finally out into an open top polyethylene
buffer tank. The buffer tank provides enough positive section head for the pumps below.
Sludge removed from the settling process is emptied to a sub-grate sludge chamber. From
this chamber, centrifugal pumps move the sludge to a cyclonic separator to remove all the
water. Once the sludge is dry, a screw conveyor deposits the material into storage containers
for removal. This dried sludge is used for landfill and as filler component for cement
manufacturing. The piping to the various tanks and process equipment within the plant is
interconnected for redundancy and concurrent maintainability. Duplex pumps are used to
pump the clarified water into sand filters. The water trickles through the beds of sand
in the filters and as further clarified in this process. The treated water is stored
in three above grade water tanks where chemical treatment takes place to control microbial
growth. The water is pumped from the canal water treatment plant to the cooling modules
via underground parallel redundant distribution pipes. The distance of 30 feet separates each
pipe and the mechanical yard is fed from both lines in a loop fashion for increased redundancy.
Samples of the canal water before and after treatment showed the effectiveness of the
water treatment plant. So, just to--to remind everyone, the Belgium datacenter is also the
chiller list datacenter. So, on a 100% evaporative cooling with a 100% reused water from an industrial
canal; we've been--been running this facility in production now, serving traffic for about
six months. Thank you very much. >> Hi there. Oh, here we go. I have a couple
of questions. One, when you say that the water is consumed, you mean evaporating goes in
the air, right? >> KAVA: I'm not--I'm not taking it into consider--mic
here--good. I'm not taking any consideration the, you know, it might be re-deposited as
in the form of rain. >> Okay, yup. And secondly, you said, you
mentioned, I believe this set of, you know, upstream, the power generation plants are
using a lot of water, presumably in a similar sort of capacity, right? Whether to use it
for cooling there… >> KAVA: Yeah, whether it's--it's once through
cooling or not, the vast majority of it ultimately returned to the waterways, but there is still
a few percent that's lost in evaporating… >> And--and that's just lost to the evaporating.
So--so, from--you mentioned then in--the none Belgium data center, you said there was, you
know, water that's self-sufficient whatever, it was using waste water from one of these
power plants? >> KAVA: No. Not from a power plant, from
the local water and sewer authority we take the basically the effluent from their water
treatment plant. >> Okay. Yeah, I wasn't sure what you were
saying there. >> KAVA: The other thing to notice is the
treatment plants don't make the water potable. So, we don't treat it to a level of potability,
we treat it to a level for industrial use where it's good enough for cooling towers
but it does not take the place of our potable water supply for our offices, or other water
on the data center. >> And my final question is, basically...
>> KAVA: I wouldn't drink that. >> Yeah. I know, I hear that. The Final question
is, you're considering that as, you know, sustainable because you don't think that will
ever dry up, you don't think that industrial canal will ever dry up? You're still using
the water, you're still evaporating the same amount of water, it's just that you are more
confident the source is that why this is a sustainable thing or could you elaborate in
that a little bit? >> Yeah. It really is this--we went to the
self-sufficient from the potable water supply used to kind of sustain the growth of population
and to sustain the growth for other users. So, we are not taking, our goal is to not
take potable water from the local authority and use that for industrial use.
>> And virtually in all communities, you know, freshwater comes in from some source, you
know, aquifer or river or, you know, whatever and gets used and gets thrown away and we
want to hang off the thrown away part so that the net increase in the community is zero
of the inflowing potable water. Right. And assuming that the community doesn't go away,
that water really will be there, you know, to be re-cleaned. So, you know, remember every
time you flush your toilet, you help us cool the data center.
>> Yes? >> TILLY: Hi, Mark Tilly from DMware. Just
a curiosity question. Most of us are still forced to look at what the ROI will be long-term
for any investment like this, how did you guys do the ROI in this particular project?
>> KAVA: So, much to the kind of the same rationale that Olivier was talking about--from
simply an ROI calculation because water today is so inexpensive, it's not compelling. It
really isn't compelling. So, it was more about doing the right thing, sustainability and
also about business continuity and guaranteeing our access to water. What if the water authority
in the area said, "We're going to cut everyone, all of our industrial users by 15% because
of severe drought conditions." By distancing ourselves from that impact, we kind of assigned
a value to that business continuity in our TCO file.
>> Totally supported just most of our business managers won't unfortunately.
>> KAVA: It will. >> TILLY: So, there are actually are geographies
where you can make a really compelling argument so, we've looked at that. One of our facilities
actually--I think the argument is, within the hair of anything that would satisfy finance
and so with a little bit of increase in the rate which is, for example, droughts happen
or more often in the US, you'd really be comfortably there, right. In other places water is incredibly
cheap and no pay back payer will ever... >> KAVA: It's a long term threat basically,
it scared the heck out it. >> HOELZLE: And really the [INDISTINCT] in
Europe as well, water is actually considerably more expensive than here and then with scale
the water plant, you know, per gallon, cost actually comes down like quite a bit. And
so, scale helps you there. And then in Belgium the situation was particularly ideal because
there was no pipeline that needed to be built. So this canal is literally in a hundred yards
from our property. And so that reduced the construction cost to that quite a bit as well.
>> TILLY: I'm not trying to take away from any either. I applaud the activity.
>> KAVA: No, you're--see the exact same type of analysis that any moment has to go through,
and certainly at scale... >> HOELZLE: Right.
>> KAVA: It makes more sense also. >> HOELZLE: Yes. And we have a project that's
not live yet that will be clearly ROI (ph) positive with Ray Motor Collection on the
site and that is very easy to justify, just purely from a cost basis, it's extremely cheap.
>> I got to ask, I saw a lot of 80% type of motors there, how are you--your Belgium (ph)
data center, how does that work out in your PUE if you sum all of that in efficient here?
>> KAVA: I'm sorry, 80% what? >> 80% PF motors, 80% Power Factor Motors
and all the, and all the--essentially you got to plan but it's not very efficient.
>> KAVA: Right, right. >> HOELZLE: So we don't... Sorry.
>> KAVA: No, I was going to say is that we don't actually calculate the PUE of that water
plant right now. It is a good point; those motors are not very efficient overall.
>> HOELZLE: But there are two reasons for not doing that. One is that everyone else
doesn't count, sort of the embodied energy of the things that are brought to the site
like water in the PUE calculation either. The second point is going to be the, in the
grand scheme of things, the power usage of that plant is, you know, epsilon, relative
to the site. Because, yes, you have bunch of pumps but they are in a hundred of horsepower
pumps kind of things. So it's hard to really make a difference. So even if we did count
it, I doubt that, you know, once we have some IT load up there it would really show up in
the stats. >> KAVA: Right. The, I mean that the entire
plant is--has it's own dedicated back-up generator as well and that is a very small portion of
the overall back-up generation that we have onsite.
>> STANLEY: So I'm John Stanley with Uptime Institute.
>> KAVA: Hi, John. >> STANLEY: And my question is--it seems like
there is kind of a high number and a low number for the water consumption of the electricity
production. The--you can get the high number based on how much do you have to draw. But
then if you only--if you multiply that by the 2% to 3%, that actually goes out in evaporation
and to consider the rest as going to return and therefore not consumed, then you'll get
a low number. So in your earlier graph, on saying that a water list data center with
the PUE of two, actually uses more electricity than efficient Google evaporated cooling data
center... >> KAVA: Uses more water.
>> STANLEY: Sorry, yes you're right, use more water. Because of the water consumed by electricity
is that--is that using the high number for electricity water consumption or the load
number? >> KAVA: That was using the--from the National
Renewable Energy labs, they report a weighted average on the US of two gallons evaporated
for every kilowatt hour of electricity produced, and that was used in that calculation.
>> STANLEY: Okay. So that actually sounds like a low number.
>> KAVA: It was the evaporation, it was not the total. It was just the waited evaporative
consumption. >> HOELZLE: Excellent. All right, one more
question. >> All right. That I would target. How about
two quick questions? The chemicals that you use for treating that water then it goes back
into the canal when you're done with that, is that--is that all environmentally friendly
chemicals that you're using? Is that...? >> KAVA: So, right--what we're doing at the
water treatment plant itself, it's only just treating for a biological growth, for microbial
growth. Once we take the water to our tanks for the cooling tower make-up water, we actually
do treat it for the normal, you know, suspend it results in solids and the rest.
>> Okay. And then on a--on a data center where you use mostly evaporative cooling but you
have chillers too, and these are off line most of the time? Do you--are you running
pumps or something to get water treatment through those chillers or...
>> KAVA: So we have a bypass loop that we take the water supply outside of the chiller
loop entirely when we're running on basically economizers.
>> Okay, got it. Thank you. >> HOELZLE: All right, thank you very much,
So Luiz is going to add some concluding thoughts and then we have a special presentation afterwards
with a topic that was not in the agenda. >> BARROSO: So I'm not going to take more
than just a couple of minutes. I just thought it would be a good idea to sort of do a quick
summation of the whole day, at least for my pair of eyes as an observer for most of it.
First, I guess on behalf of the Google, I want to thank you all for coming here. I think
that we have--because of your participation and particularly because of the presenters
and panelists that we have here that are not from Google, in event that is much more interesting
then sort of Google disclosing sort of some detail. And I think that this made it so much
more interesting for all of us, so I'd like to thank you for this. I wanted to, you know,
again, you know, mention that you met some of the folks that are responsible here for
all these new disclosures, some of them that I think you would agree they are quite interesting.
A lot of them are here in the room and you've been able to talk to them and a lot of them
are not here as well, this is a work of a very, very large organization including, you
know, of Data Center Operations and Platforms design team. So, I wanted to make sure to
sort of remind us all that that this is sort of an effort for a large company. Some of
you guys actually gave us credit for breaking organizational barriers in the way that we
find solutions for our problems that are really sort of looking at the hard scope of things.
And I want you to just embarrass Urs here then for a moment and claim that, you know,
Urs Hoelzle is largely to blame for this. You know, he have been a software guy, in
particular, a compiler guy, running you operations is probably a good thing in terms of aligning
incentives across an entire company. I wanted to close then with just summarizing a few
thoughts, that's so, to just keep coming back throughout the day. One of them is that clearly
I think you agree with some of the comments we've made earlier that a lot of the improvement
that you've seen in our data centers that were detailed here for the first time are
actually, you know, achievable by a lot of people, perhaps everyone. Maybe not down to
the percentage levels that went to but way better than the sort of industry averages
and we're hoping that the kind of disclosure that we did today will sort of help create
incentives and targets that are more aggressive for people to do that. Some of us mentioned
or of you asked earlier why is it that we're talking about these things right now, I just
to repeat what Jimmy Clidaras said a little bit earlier today, its way more fun to talk
about this than it is to keep things secret. So trust me, we would like to be able to disclose
as much as possible. We are cognizant of the fact that good solutions have to have a strong
economic impact like Ken sort of mixed as a point of reminding us and has to make sense
with businesses. These are things that are actually done by people who are idealists
as well, who are sort of wanting to do good things too and these things actually are so
fairly well aligned. We will have run across the whole business that happened to be able
to do some of good things into efficiency. And you know, disclosing that they, kind of
data with you today I think it's one of the main motivations is that, our work with EPA
has been made very clear to us how important it is to show, you know, existence, proofs
of this, right? And I hope that you take that in that spirit. But basically the solutions
are the ones that are sort of all of you know sort of very well, the bag of tricks is the
same thing, right. You sort of understand climate effects. You control sort of air flow
as much as you can. You try to use the chillers as little as possible, for example, you pay
attention to water usage. You look at the whole picture. The observation that you're
also going to make just now I think is a good example of that. You know, if you look at
just your consumption of energy and you look at water usage, some solutions may look better
than the others. And when you look at the entire pipeline where the electricity come
from, the situation is a little bit more complex, right. Then, so looking broadly at problems
is something that kept coming on and on throughout the day that I think was useful to think about.
Pushing our equipment a little bit warmer, right? James talked about this, something
that as you noticed in our design that we have taken advantage of. And it turns out
that you actually look a little bit in terms of equipment we had at the paper awhile back,
you look at feather weights (ph) for disk drives for example that some of you may have
seen. That actually looks at an entire fleet of disk drives and asks yourselves the questions,
what happens if you run these things five degrees a little bit warmer? Is the sky going
to fall? And it turns out that it doesn't, right? And so, there are certainly opportunities
that we need to take advantage of that are on disk space. And then, finally, Olivier
have mentioned this at least a couple of times that I think it's important for me as a software
engineer today, go back to doing some of the hardware and software side of things. I need
to mention this, too. These big machines are running software, right. The useful work that
they do is the software that's written on top of them. And the best facility and the
best management upgraded facility that's running crappy software and is not doing anything
for you. It's not going to be a very efficient one regardless of how good that PUE is. And
it has to be possible. It has to be something that's possible for engineers with not a tremendous
amount of efforts to write programs that are energy efficient on your facility too. So
part of delivering that whole platform is not creating a system that while you can save
energy but only two programmers in the whole world can design the software that's going
to save energy on that facility. Those are not acceptable solutions. So I wanted to close
and look with that and pass the baton to Urs. He's going to tell us about some of the exciting
project. >> HOELZLE: Thank you, Luiz. So, we have talked
a lot about what we've done. And many things that we've talked about are things that already
have been done, you know, that were done five, even six years ago. And I thought it would
be good to conclude by giving you a little bit of insight and something that we're doing
right now. So, literally, this week people are working on it. And the topic is Oil-Cooled
Data Centers. Now, you may think, you know, what the heck, right? But it's not actually
that crazy an idea. So, oil has been around for a long time for cooling transformers,
right, and car engines. Oil has great properties, it's not corrosive. It doesn't conduct electricity
so it can imbed stuff in it; non-reactive. The problem is that it's really inconvenient
to use, and that's why nobody has really used it. And that's why also for a long time we
haven't really thought about this, because the oil is actually relatively expensive if
you needed that volume. Containing the oil is even more expensive, right. And then to
add insult to injury, if you try to do it , you know, the AHA content says, "Actually,
you need a major permit and it's going to take you five years to get that permit," if
you actually want to do that, right. You all know that from images and structures. It turns
out they're actually our answers to these problems. And the answer are oil tankers,
right? Oil tankers are built, custom-built to hold oil. And the Economist actually just
had an article 10 days ago about the cost of shipping drastically collapsing in the
past few months. So in the last, literally since last summer, basically the market doesn't
exist anymore because all these ships that have been shipping oils from China or oil
from Saudi Arabia really are not very busy anymore. It's equally easy to park the ships
somewhere. And, you know, we all heard the rumors about barges on the ocean. One of the
problems with having things on the ocean is that your power and your connectivity is far
away. If you actually have the boat in harbor, these problems go away. The engines still
are very useful. These ships have powerful engines and, you know, all it needs is an
alternator for them. And then finally you don't need to pay for the oil. You can basically
rent the oil. There's many place. If you go to any major harbor facility, you'd see these
all storage tanks next to it. Oil flow, you know, oil isn't really stored in there. Oil
flows through there. And so you can get with the right contractual arrangement someone
to flow oil through your container, your ship instead of one of these bay tanks. Now, it
still sounds crazy and I'm going to show you why actually it isn't. So here's an overview
of a typical tanker, you know, very simplified. And a tanker basically is, you know, a hull
and a bunch of individual oil tanks in it. There's really nothing very interesting. It's
very simple. That's why actually the $70 million that I mentioned is literally a scrap value
of the ship and you can go look it up in the Economist. These ships are selling for scrap
value today. Try to build this oil, you know, this kind of volume yourself. Even in a warehouse,
you're going to spend more than $70 million today. So, it's a very attractive housing.
Now, what can you do with it? Well, very simply. You take the tanks. You put your containers
in those tanks. I don't know what's so funny about that. So you don't fill the tanks fully
because all it needs is to get the convection cooling going on that oil. So really the conversion
cost of that is very, very low literally to anchor the—volt down the tanks—the containers,
sorry, supply power and you're basically done expect for the UPS conversion. And the really
cool thing about that is even compared to our efficient facilities; you eliminate a
whole bunch of stuff. There's no air moving. There's no air handling. There's no coils.
There's no pumps. There's no water. There's really--everything basically disappears and
convection cooling takes care of it because these ships are very, very poorly insulated.
Nobody pays any attention to insulation, of course, and so as a result, you just dissipate
that heat into the ocean. So we actually bought one of these ships about three years ago--three
months ago. The logo is fake. It's not actually painted there but we felt that was important
to have it there. And we invite you to join us on a tour of the ship. That's moored just
about 10 miles from here. Buses are leaving after our reception today and yes, no pictures
please. You need a special ticket. You can get it on the way out actually from people
wearing red shirts like this one. You know, on the seats, so, there should be plenty for
everyone but please get a ticket before you line up. We are going to have a reception
now so that people can mingle. It's going to be where our lunch was so we can kind of
wander over there and have some drinks cocktails, appetizers and things like that, and so, get
your ticket on the way. Thanks again for coming again and we'll stick around to talk you about