Slashdot Mirror
Hot Aisle Or Cold Aisle For Containment?
1sockchuck writes "Separating the hot and cold air in a data center is one of the keys to improving energy efficiency. But containment systems don't have to be fancy or expensive, as Google showed in a presentation Thursday, which discussed the use of clear vinyl curtains in isolating hot and cold aisles. Containment systems have been in use at least since 2004, but there's an ongoing debate about whether it is best to contain the hot aisle or cold aisle. Leading vendors are split as well, as APC advances hot aisle containment while Emerson/Liebert champions a cold aisle approach. What say Slashdot readers? Do you use containment in your data center? If so, do you contain the hot aisle or cold aisle?"
Does that count as cold aisle containment?
I thought this article was about supermarkets, where it might be a good idea anyway too..
I leave this on: http://home-solutions.hsn.com/amana-12-000-btu-portable-air-conditioner-and-dehumidifier-with-remote_pf-450623_xp.aspx?club_id=450623&sz=888&sf=HW0130&rdr=1&cm_mmc=Shopping%20Engine-_-Froogle-_-Home%20Solutions-_-5502594&cawelaid=349282299 ... and make sure the closet door stays closed.
Depending on how your facility is ducted, it might not cost much to try both options and measure the results. Even if you have to spend a few thousand doing so, the long term savings from choosing the best method for your site would probably be well worth the cost of testing.
What say Slashdot readers? Do you use containment in your data center? If so, do you contain the hot aisle or cold aisle?
I think that I speak for most readers here when I say that it's pretty much all cold aisle down here in my mom's dank basement. Not much containment either, other than some pegboard partitions.
Contain and exhaust your heated air, vent it up outside
That way it doesn't mix with the cold air much.
If you just contain your cold air, then you have a situation where the hot air is staying in the room, and that heat will be absorbed over a larger surface area, by all the things in your server room (including the Air handling units).
that I would like to have.
Once your server room is to the point where you have hot and cold aisle, just contain one and go for a beer.
20 characters max for the password? How will I use my favorite poems as passwords?
We only resort to using containment when the servers have been very, very naughty. We've found that chains, steel cable and duct tape are the best ways to keep servers in their racks.
If "disco" means "I learn" in Latin, does "discothèque" mean "I learn technology"?
If we were to retro-fit it at work, I'd say cold aisle. To do so would mean curtains at the end of the aisles, as the under-floor vent grids are in front of the racks. The CRACs are at the end of the room sucking in air through the top, so it'd be cool air pumped up through the floor, into a cold-only zone, sucked through the racks, blown out the back into the rest of the room where it just swirls about until it's pulled into the CRACs again. I reckon it could be done cheaply and quickly. Do do it with the hot aisles would require more containment to get the air back to the CRACs. I think it'd be a case of which air flow it fits best.
What did I find when I joined the sysadmin team at my place?
Putting cold air vents behind the racks doesn't help. Pull cold air through the front to the back? Nope. Chill the exhausted air because it sucks to walk behind the servers. Nice.
My mom says I'm cool.
Patent keeping the not side hot and the cold side cold.
Maybe I'm missing something obvious, but the answer shouldn't be complex. Base the decision to contain either hot or cold aisles on the differences to ambient temperature. If (HotT - AmbientT) > (AmbientT - ColdT), then contain the hot aisles. If it's the other way around, contain the cold aisles. This minimizes the entropy loss due to temperature mixing in the data center, I believe. Just my 2 cents.
sigfault (core dumped)
The best solution is going to based on the average ambient temperature of your location. If you're in a hot environment, why contain the cold if you need additional A/C in the datacenter for employees? Reduce costs by using the same equipment to cool both. If you're in a cold region, then let the heat also warm the datacenter. If you're in an ideal temperature environment, then you don't have much to worry about beside good air flow.
One critical thing is where are the HVAC return ducts and where the air vents are. Does the datacenter use raised flooring, or does the place have discrete ducts for its ventilation. I've seen data centers have 12-24" of clearance used as plenum space. Others may have 2-4 inches because the space is used just for wiring and not HVAC use.
This needs to be factored in to separate the aisles, elsewise spending time for meat locker curtains and endcaps like Google has done may bring few to no returns.
Whatever one Apple uses, that's obviously the evil one.
Why is this modded offtopic? OP was clearly stating expressing his support for hot aisles.
I've been in quite a few large datacenters. Some have strict rules on proper utilizing their hot and cold aisles. Some could care less.
The ones with the best ventilation have the cold air coming through the raised floor, and the hot air being pulled from the ceiling. Brilliant. Actually understanding that hot air rises. :)
Most
Some I've been in had the hot air being blown from the ceiling, and the return somewhere on a vertical wall.
At one place, they worked with us on it. We had two rows in a cage. We established the center to be the cold, and the edges to be hot. It wasn't really for temperature, it was for access. We spread the rows a little extra so we could have a couple carts and two people working at the same time. In that, we couldn't load some of the longer machines in from the "hot" side if we had wanted to. So on the aisle that we worked in, they gave us more cold air outlets, and sealed off the ones on the hot side. It worked very well. The site manager was really into making everything work as well as possible. He would walk around with a non-contact IR thermometer and spot check equipment.
Separating the hot and cold aisles with plastic would probably never work. Most racks that I've seen either are open frame, or they have a vent fan in the top. The open frame ones are obviously worthless for a plastic barrier.
Serious? Seriousness is well above my pay grade.
I can honestly say you win either way. The electricity/cost savings of containment will pay for itself regardless of where you put the doors. That said, whether you choose to go HAC or CAC is really choosing between different trade-offs.
HAC (The APC method): Seemed to be cheaper and easier to install. Since the hot aisle is being contained, if something happens to your coolers, you have a longer ride-through time as there's a much larger volume of cold air to draw from. However, at least when I got out of the business, HAC *required* the use of in-row cooling, and with APC, that meant water in your rows. Europeans don't seem to mind that, but Americans do (which provided an opening for Emerson's XD phase-change systems, dunno if APC has an equivalent or not yet). I personally wouldn't be too keen on having to spend more than a few minutes inside that hot aisle, either.
CAC (The Emerson method): Seemed to be more expensive, especially in refit scenarios (they appeared to be more focused on winning the big "green-field" jobs more than upgrading old sites), but it can usually leverage existing CRAC units, so you could potentially save enough there to make it competitive, as well as avoid vendor lock-in. The whole room becomes the equivalent of a hot aisle, but convection and the building's HVAC can somewhat mitigate that, so it'll still be uncomfortable working behind a rack, it doesn't feel quite the sauna that an HAC system does. Depending on whose CRAC equipment you buy (or already have), EC plug fans and VSD-driven blowers can save even more money if properly configured.
Other: I've seen the "Tower of Cool" or "chimney" style system, and flat out hate it. They look like a great idea on the face of it: much cheaper, faster installation, able to use building HVAC, etc. But let's be honest. Your servers are designed for front-to-rear airflow. So are the SANs, NASs, TBUs, rack UPSs, and practically everything else you've put in your datacenter, apart from those screwball Cisco routers that have a side-to-side pattern (Seriously... what WERE they thinking on that one???). Why would you then try to establish an upwards-pointed airflow that's got a giant suction hose at the center of the rack's roof, where it can just as easily pull cold air from the front (starving your systems) as it does hot air from the back?
Personally, I like cold aisle better. If I'm going to be spending two hours sitting behind a server because I can't do something via remote (forced into untangling the network cable rat's nest, perhaps), I like the idea of being merely uncomfortable and a bit sweaty than dripping buckets while cursing the bean-counters who forced me to lay off the PFY two months ago. There are also some neat controllers that work with CRAC units to establish just the right amount of airflow to fully feed the row and manage their output, so if running five CRACs at 50% is more power efficient than running three at 100%, that's what they do. I know folks who like hot aisle better. It's more fun for them to show-off their prize datacenter since all the areas you'd want to see (unless you're the one responsible for power strips or cable management) are cool.
I suspect that one of these choices is incorrect. Correct.
It is pretty basic. You are paying for the cooling, and you don't care if the heat bleeds away, but you care if the cold warms before it does its job, so you partition and direct the cold air and let the hot go wherever. The bottom line here is the hand waver types have never done a cost analysis based on required cooling loads, which is what the data center is paying for.
It really doesn't matter which side you contain as the end result is the same.
The important part is preventing the cold incoming air from being tainted with hot exhaust air. Which side you do it on is largely irrelevant unless you have a large amount of foot traffic in the server room which may bring in outside air to upset the balance, but even that would require what I would most certainly consider an unacceptable amount of foot traffic to occur in a datacenter.
Just hanging vinyl like Google does takes you to the point that you're wasting time and money by putting any further engineering resources on it.
Persistent Volume manager for Kubernetes - https://github.com/dwimsey/openshift-pvmanager
SLAs are usually based on the temperature of the cold aisle. Containing the cold aisle along with proper blanking of cabinets and equipment venting the right way would help to keep within the contract.
When you contain the heat, you then have the ability to move it around and use it for cogeneration, thus vastly increasing your overall efficiency.
The answer will be specific to each implementation.
But in general, it should be pretty obvious to anyone that understands basic thermodynamics: get the "cold" into the servers without mixing it with the ambient or letting it touch any hot metal, and get the heat out of the servers without mixing it with the ambient or letting it heat up any other metal.
It should be pretty obvious that air is not really the best way to do this; air goes all over the place, and is not a very good thermal conductor (relatively speaking).
There are entire 10k+ machine datacenters in France that use only liquid cooling circuits, right up to the servers. Energy costs for running the external condensers are a small fraction of what it would cost to do the same thing with air. Of course, it helps if you only have your own machines in such an environment, but if APC, Emerson, etc were serious about efficient cooling then they'd partner with HP, Dell, etc. to make standardized systems that would allow this...
Hi There,
We did some testing on two rows of 27 racks which we had fully populated with rack mount servers. We measured each setup over a two month period during a european summer. We found that by containing the cold rows we saved around 20% on the power bills overall. For the hot isle containment we found that we saved only 12%. Unfortunately for us the testing that we did was not as comprehensive or detailed as it should have been. I believe that there are so many other factors which affect the overall power saving. For example some vendors recommended having the air con in the racks while others say that its best to have the air con outside of the solution with pressure forcing the air through the floor. We didn't have the opportunity to test the air con in the rack because it was a retro fit in an existing facility. Overall we found that the isolation was a good thing and saved us the cost of installing within a few months. Right now we have 18 rows of 27 racks with cold air containment working flawlessly. Some other points which need to be noted..
- a squared off ceiling over the cold rows does not provide good enough air flow. Having a curved ceiling was good.
- ensure all empty rack slots are filled with blanking panels.
- make sure that the gap between racks is filled.
- make sure the exhaust fans on the top of racks are only on the back of the rack not at the front like in some installs.
- make sure that the cabling is done in a tidy way, any air obstruction makes a difference to the efficiency. we got special power cables made up so that there was no slack hanging around.
Over the 18 rows we have now had the install running for two years and over that time we have saved an average of 28% on the power in that facility. Which when your talking about 30+ servers per rack with 27 racks in a row and 18 rows it makes a big difference.
how about really high ceilings in a funnel shape and vents that you can open to the outside (with filters). Most data centre rooms i've seen are normal office height. after all let the hot air rise.
While we are at it let's turn the racks and the pizza boxes around 90 degrees - less distance for air to travel over hot components, likley require lower fan revolutions as well. Not to mention much hardware doesn;t need all that depth and with more 'front' area we could fit more swapable hard drives in per U for the virtual machines either with single servers or SANs.
anyway just a thought.
I've seen the chimney method too, with a large fan at the top. Nice in theory, but I have seen almost no rack units made for bottom to top cooling. Almost everything goes front to back. Ironically, one place I worked at received a ton of racks with the stupid fans on top. Of course, the fans were yanked, and a piece of metal screwed on top to ensure the airflow went the proper way.
I wonder what the future of this stuff will be. It would be most efficient to replace CRACs with water chillers and having a liquid cooling system that doesn't just cool the rack, but has heat exchangers and systems to cool individual components (down to the RAM, CPUs, hard disks, and other items.) However, more sophistication is needed to be done with valves and the ability to shut off or shunt subsections, all it takes is one bad hose, and the data center suffers both water damage as well as heat failure. Until someone invents not just a valve assembly that doesn't leak after hundreds to tens of thousands of insertion/removal cycles, as well as an assembly that can detect water pressure loss and close off/shunt, liquid cooling will be only usable for systems that are essentially static and where devices are rarely connected/disconnected from, such as HVAC systems, APS racks, or PC systems.
all the air in a machineroom is either hot or cold. anything else means you're mixing - that is, your containment leaks. there is basically no heat transfer through building conduction, for instance. 'cold' merely means that it's between the chiller outflow and front of servers; hot means ass-side of servers and chiller intake. the primary goal is to keep them from mixing.
a nontrivial machineroom will have multiple chillers and non-uniform heatload distribution. that doesn't change this principle, but does mean that the airflow design may have trouble getting enough air to the right spots. ideally, the chiller outflow would all go into a single large plenum (such as a deep pressurized underfloor), with hot ducting with controllable air intakes whisking the hot air back to the chillers.
so contain the heat so that it doesn't encroach on the cold.
My limited understanding of thermal dynamics is that the more active molecules
in a heated environment tend to move toward cold which is a lower energy state,
thus entropy plays an ever present role. Contain the heat, prevent the transmission
of infrared and use any other barrier you can to prevent heat from warming the
cooler side.
There are a lot of reasons why someone will be sitting in a server room for an hour or more. Please don't make it an unbearable hour with heat baking the poor humans.
Depends on the climate entirely. Here the summers are brutal and the winters severe. which one is contained is pretty much up to which one we want to be exposed to. (i.e. winter it is nice to have the heat NOT just dumped outside.) Curtains and thought to the existing doors/partitions help with that seasonal flexibility a lot.
I mean it worked for the McDonald's McDLT back in the 80's...
'He who has to break a thing to find out what it is, has left the path of wisdom.' -- Gandalf to Saruman
CYA. Flip a coin. That way, if you get complaints about the wrong choice, you can always blame the coin.
it makes for hard nipples on your staff
First, what power density are you planning to run? You'll find the debate ends about 15 to 20kw. Then its all hot containment or in-row or back of cabinet.
This basically comes down to a question of APC vs Liebert equipment. APC pushes hot aisle containment where Leibert pushes cold aisle containment. It seems like the benefits are almost split 50/50, but there are a few other things to consider. The APC method uses in-row cooling, which means you are moving the the air less distance than the Liebert method of pushing air from large CRACs at the ends of the aisles. There are efficiencies gained in other methods, and as a poster earlier mentioned the APC method is a little cheaper up front. This does introduce the problem of water in the data center, but for us we resolved this by putting the water under the floor so that it would be very difficult to have any issues related to it. Overall, I think this comes down to the situation, but for us it made more sense to do hot aisle containment. It seems to be better in smaller scale situations, but I would have to do more research to see how it scales to large 50,000+ sqft data centers.
After that, precisely how you go about the process is more a matter of the side effect of other choices.
This feels, to me, rather like arguing if it's better to put the women's washroom to the right or left of the men's.
Sometimes boldness is in fashion. Sometimes only the brave will be bold.
Then it’s easy: Do the math. Seriously. It’s a question of energy transfer, the simplest thermodynamics, and isolation.
I’m not a expert in those matters. But I’d say, the goal would be, to minimize energy (heat) differences between areas right next to each other?
I’m sure there is an expert (please no wannabes) here who can quickly give a nice answer. :)
Any sufficiently advanced intelligence is indistinguishable from stupidity.
Hot aisle containment is more efficient, as you are extracting heat from a smaller volume of air. However, remember that in a hot aisle containment system, if you ever have to work on the back of the rack, you are walking into a dangerously hot environment. The trade off is between efficiency and safety, IMO.
We have to choose between the glass being half full or half empty. But it's not the symmetrical choice which it might seem to be. Specifically, it's not a matter of providing cold air. That's just a means to an end. Fundamentally, it's a matter of removing thermal pollution.
The ideal environment for the equipment is one which is uniformly, ambiently cold. Not only are there fewer thermal stresses, the entire design problem is simplified if you can assume uniformity. Departures from this ideal are therefore to be minimized. You don't want to contain the cold, you want it to prevail. Instead you want to contain the heat, remove it, and minimize points of contamination along the way. This has the additional benefit that you're minimizing leakage from the relatively small heat regions rather than trying to protect the entire environment.
An analogous situation arises in a marine engine room. You want to supply clean air and also exhaust the combustion gases. Supposing you only had the following two choices, should you (a) exhaust directly into the engine room and pipe fresh air from outside to the engine, or (b) pipe the exhaust outside and let air be drawn directly into the engine? Generally (b) is regarded as better, because exhaust gas is hot, toxic, and corrosive. It's much more trouble to design every surface in the environment to tolerate it that it would be just to contain and get rid of it.
Parity: What to do when the weekend comes.
It's cheaper to generate heat than cold. So I'd go for cold containment.
Actually containing both might be best since then you will have a "room temp" air gap between the two and air is a fantastic insulator. IF you do not contain the hot then the heat will diffuse and the air on the other side of the vinyl curtain will be warmer than room temp. This will warm your incoming cool air. The effect may not be particularly noticeable but it would be an interesting test to see if there is a noticeable improvement to doing both.
The -1 mod for above is a prime example of why I don't bother reading /. anymore most of the time. No, it's not redundant. Check the timestamps and check others who got insightful for later posts.
That's exactly what the server designers should be doing to the rack series servers at this point. Stop with the loud and inefficient air fans and replace them with built-in thermal conduction pipes inside the servers. Every rack should start coming with a master hose and a coupler system that we would connect to the server, prime the server before turning on for the first time, and let the server be cooled with liquid and have the cooling capability go directly to server parts that need the most cooling. The flexible but reinforced pipe system to each server would use self sealing ball-bearing couplers to avoid spillage on connect. The thermal substance could be your ordinary Ethylene Glycol or something better.
This is already being done in specialized circumstances and we have been doing water based cooling on our overclocked rigs at home for quite some time, it's only logical that the servers start moving down this path and we get liquid cooling into systems that can absorb a little bit of a price increase in exchange for lower overall energy costs.
Many of the data centers in NYC where I've worked already use a Ethylene Glycol based Chilled Water cooling system so this technology already exists in these locations. It's only logical to make some servers with liquid cooling components for the hard drives, power supplies, memory, processor, chipset, or just borrow these pre-made parts from liquid cooler overclocking manufacturers. Set up some pipes, put in some safety low-pressure valves, self-sealing couplers, and run them to all the racks over the top like network wires.
There are more than enough engineers smart enough to get this designed and organized and this is the right time to do it with Green Computing being such a fad at this point. Someone wake some people up at HP & Dell.
I am an HVAC designer by trade, and I think that a hot isle containment setup in general is a good idea. especially if you are trying to use something like curtains as your containment. when you have a standard CRAC unit, they have been until recently only constant airflow volume units. but the servers themselves have their own fans that stage or change speeds to move the required air through the racks. unless the airflow from the CRAC is being varied based on demand, the extra airflow will just spill out past your curtains, out the ends of the isles and cool the room outside the cold isles no matter what you do. newer approaches (including the Liebert approach) advocate sensors on the racks to calibrate demand, and do match airflow to the load. In these cases, containment is not strictly necessary, because the racks are transferring nearly the same air through from hot to cold as the unit is supplying and returning. the natural air convection serves to separate streams pretty well. but, you could gain some efficiency on the CRAC unit cooling side by containing the hot isle air, the idea being to bring back the warmest air possible. obviously the air mixing with cold air is a source of wasted energy, but also, the refrigerant cycle can operate at a warmer relative temperature for a given amount of cooling. when you require the cooling coil to operate colder, the cycle efficiency drops off. Of course, there is also energy saved by not operating the fan itself at such high volume. but, for my money, variable air volume and proper controls to match airflow at the unit is worth more than trying to set up containment, especially since the curtains or other devices make for a less flexible layout, while relocating sensors to critical racks (the ones that get hot first) can be done quicker.
You've hit on an important point here: human beings do in fact do some work in this space. Do you really want the guy racking your servers to have wet, sweaty palms?
Moderating "-1, Disagree" is simple censorship. Have the guts to post your opinion.
Flexible barrier -> contains high pressure -> cold (or it would be ineffective).
Rigid barrier -> equally applicable to high and low pressure -> hot (more convenient, can be used with varied fans speed).
Contrary to the popular belief, there indeed is no God.
At my place of employment, we do something similar to http://www.42u.com/images/great-lakes-exhaust-chimney-diagram.jpg. There is no hot air anywhere outside of the cab.
We just built out 152 cabinets of Cold Aisle Containment. Think about this:
4,000 Sq Ft Room
8 x 30 Ton Liebert CW Downflow
36" Raised Floor with no obstructions in Cold Aisles and only 10" network cable trays under the hot aisles (power overhead)
Self Engineered CAC Ceilings 12" above tops of cabinets
Self Engineered CAC doors at the ends of 4 Cold Aisles (18 cabinets on each side of the cold aisle)
9' plenum above the dropped ceiling
CRAC units ducted to the plenum for hot air return
Here is the MOST important thing to think about conceptually. We are only actually chilling 38'x4'=152 Sq. Ft per cold aisle x 4 ColdAisles = 608 Sq Ft of actual datacenter space with 8 x 30 Ton CRAC units... This should be basically as directly efficient as the people that put cooling directly into the cabinets.
It will be a bit strange to walk into a datacenter that runs at 80-85 degrees F ambient (until you go into the cold aisle and shiver.
There are several factors which, in my opinion, must be considered when making a decision on what to do:
* Layout
Both hot- and cold-aisle containment require that you have either one row of racks, with a wall on the other side; or you have two rows of racks, both of which are of identical length and are aligned properly. This makes a long, thin room better for hot- and cold-aisle containment.
* Raised-floor
Is your environment going to have a raised floor, or (probably the more apt question) does it already? If you go for hot-aisle containment with a raised floor, you'll going to have an interesting time getting the hot air returned to the air handlers units. I think you'll also need a way to ensure air circulated in the under-floor area.
* External heat sources
Lighting and people impose an additional heat load, for which you must account. The same applies to the UPS. Also, don't forget about all of your networking equipment and patch panels that prefer living in two-post racks.
If I remember hot-aisle continment theory, the cold air being ejected from the in-row air handlers doesn't get very far into the room before it is pulled in to the servers in neighbor racks. If that is the case, and your room is big enough -- with enough space between rows and stand-alone equipment -- then you might end up with a hot-aisle-containment environment where the stand-alone equipment sits in a little hot spot, unable to draw any cool air from (relatively) far away in-row air handlers.
* Cooling Source
Most of the time, when you think of a data center, you think chilled water. That is not always the case, and I'm not talking about this newfangled stuff like using outside air. I mean some buildings simply don't have the capacity (in whatever way you want) to support the chilled-water needs of a data center. You might have to rely on an air-cooled solution.
* Humidity
You have to maintain humidity as well as temperature, and the humidification/dehumidification equipment takes up alot of space. When you're going for hot-aisle containment, you're likely going to use in-row air handlers. You want these to be as compact as possible, so plan on getting a separate unit to handle the humidity.
If you're doing cold-aisle containment, you're likely going to use stand-alone air handlers, which will include humidity control already.
All of the thoughts above were going through my head in the last 12 months. In the company where I'm working, at the start of 2010, it was looking like we were going to move into a new building. At the time, the building was an empty shell, so a data center was designed that was long (to get two aligned rows), thin (so no areas for hot spots to build up), and with no raised floor (to save the expense). We were going to go for hot-aisle containment, in-row cooling, in-row UPS, and a little stand-alone device for humidity control and backup cooling. The move was cancelled, so we started over. Besides, we learned towards the end that it would be air-cooled, not chilled water (in other words, bad).
The new plan would be to stay in our current building, extending part of our existing data center (standard raised floor, stand-alone chilled water cooling) into a space that would still give us two long rows, but the raised floor and chilled water availability -- along with the ability to re-use existing air handlers -- make me want to go cold-aisle containment. Unfortunately, that was scrapped as well.
In the end, it looks like we'll simply be taking down some existing walls in our data center, giving us several rows of racks, all of which are not aligned. Hot- and cold-aisle containment now looks impractical, unfortunately, but I did learn alot while doing my research. This move will also give me the ability to change out the hodepodge of racks to newer, deeper racks (one thing that can really help is to keep cables out of the way of the back of the server), and to install blanking panels over every open rack U (again, another easy way to keep hot and cold air mixing). I'll also be asking if it will if some of the ceiling tiles are changed out, replacing solid tiles with grates, allowing more hot air to float up into the plenum.
Good luck!
Just a note about those silly Cisco switches:
Servers have holes in the front and acbk to facilitate cooling. They can do this because the boards can be oriented in a way to facilitate this.
Cisco Rack enclosures have high-density blades in the front(no roon to breathe) and a sizeable backbone in the back (A wall of PCB).
Due to the hotplug nature of the blades, the backbone has to be mounted at the back (instead of using riser boards like in computers). The only other way to have it at the side is by making the server open at the side, like HP9000 computers. This means that every HP9000 rack has to stand alone to facilitate removal of the side panels (the servers are too heavy to slide out). You do not WANT be able to slide a Cisco 6500 out, because it usually has hundreds of cables attached.
The blades, being solid in form, do not facilitate vertical cooling, either.
This design leaves only one possible directioon of airflow: horizontal.
And, yes, it has cost a friend of mine a core-switch when the cooling gave out (there were 3 of them stacked side-by-side).
"I was in love with a beautiful blonde once, dear. She drove me to drink. It's the one thing I am indebted to her for."
H/C containment is basically a way of only having to cool half the floor area of the DC. We basically take that a step further by forcing the air from the floor into the racks from below and duct it up to the front of the racks. We don't give a shit what the ambient temp is, only that the air temp at the top of the rack in front of the server at 42U is 22 C. If the racks get too hot you use an exhaust system to suck hot air out of the racks which in turn draws more air from the plenum. Its cheaper to cool the sealed racks than cool the whole friggin DC.
Air conditioners work more effeciently w/ hotter air. If you contain the hot aisle and use it as input for your AC and have the cool air exhausted from the AC units into the cold area of the room you can reduce the power draw required for cooling.
APC Patented their hot-aisle containment, and that is the only reason they use it. While it is silly to think in terms of containing one or the other-- it is just separation-- you ideally want a high comportment for the hot aisle for stratification and stack effect. This prevents re-circulation and short-circuiting of hot air, and also makes the hot aisle a more bearable place to stand.
It's always easier to keep the heat away from the cold the smaller the containment area is. So if the hot aisle is small then contain it, if the cold aisle is small then contain the cold aisle. The smaller the aisle, the smaller the walls of the containment, the lesser the surface area, the more effective the insulation.
I've never seen machines that do this (although liquid cooling comes closest)
But what about having air ducts inside of computer boxes? (Possibly in and out)
No fans within the machine itself, possibly having the "in" duct Y off, directing the cold air directly on the CPU heatsink, after chilling the box, on the way out (when air is partially warmed) have it vent through the power supply (which needs less cooling anyway)
Then, chill only the equipment via a centralized fan, as a bonus, you can filter the air to virtually eliminate dust and never again have to repair a CPU fan.
The McDonald's LT1 Supertasty Computer system (MC-DLT) has had isolation of hot and cold sections since the mid 1980's. They used less environmentally friendly poly-styrene, as the separator, not vinyl like Google does, but that's hardly relevant.
Correct me if I'm wrong here, but if you isolate one: don't you also isolate the other (kinda the point of isolation)? Then consider the physics of a server room in terms of airflow. You have cold air being pumped to the front of the servers as fast as possible to maintain a temperature (creating a high pressure zone). that air is being passed through the servers and the now hot air is being vented out to be chilled (creating a low pressure zone). So you pass from one aisle to the next you will have some clod air pass to the hot side, but no hot air will pass to the cold side because of the pressure difference.
~ Normality is merely the achievement of the mediocre...
What was Cisco thinking? Well, they weren't thinking about cooling.... probably about who to acquire next and call it "innovation".
Never ascribe to malice that which can adequately be explained by tenure.
If your coworker is in the hot isle. My coworker does this to me all the time. I'll hear him say "Sorry!" and then a hot fart gets blown into my face.
We use a ceiling plenum ducted down to the AC units with an grated ceiling in the hot rows to create a vaccuum effect pulling the hot air in, and a 2 foot plenum underfloor with perf tiles in the fronts of the racks to deliver the cold. It's pretty efficient. We don't get a whole lot of spillover, even without damming the rows appart. We did use blanking panels in the racks to stop the convection, but for the most part we've found that air balancing works pretty well. A few small tweaks to the AC settings will give you a fair bit of control as to how much cold air goes where without too much hassle.
We are testing out the chimney idea to separate even more, but so far I'm not loving it. Could be because the damn vendors keep trying to sell me fan controlled chimneys and I keep explaining to them that if you have a vaccuum on one side and a high pressure region on the other and you want air to go from the high pressure zone up to the vaccuum zone, you don't need a fan to do it. The laws of physics and thermodynamics still work. They don't grasp it. I dumb it down to the point of "this sucks hot air out. this blows cold air in. let the air handlers do their job." Nope. Please buy our fancy fans to do what nature will do anyways. Um, no. Bad vendor. No cookie. Stupid ASHRAE.
http://switchnap.com/pages/tech-specs/thermal-scif.php for several years they have been using what they call "T-scif" (thermal separate compartment in facility) to do what google is showing here for us (sun/oracle) for many years.
It appears on the page lower in the default view therefore it shows up "after" other posts therefore making it redundant. The moderator obviously views the time space relationship from a quantum or relativity perspective and not a Newtonian notion that time is absolute.
Tricky one really – should we separate the hot air from the cold air OR (radically) the cold air from the hot air?
How about this for a radical idea – turn off the servers they’re not using until they pay the ransom?
www.rackface.com