Slashdot Mirror
Raised Flooring Obsolete or Not?
mstansberry writes "In part three of a series on the price of power in the data center, experts debate the merits of raised flooring. It's been around for years, but the original raised floors weren't designed to handle the air flow people are trying to get out them today. Some say it isn't practical to expect air to make several ninety-degree turns and actually get to where it's supposed to go. Is cooling with raised floors the most efficient option?"
...but in lowered walling.
Where else am I going to store my beer so it can stay cold and the boss not find it?
But then where will we keep the bodies?
As long as the space under the floor has a negative or positive atmosphere I can't see how somme turns have anything to do with the air flow.
Says that raised floors may be inefficient if it gets block. Then says alternatives are expensive. Direct AC where you need it, the article says.
Why wouldn't raised floors be bad if you used them properly?
A NYC lawyer blogs. http://www.chuangblog.com/
I thought the raised flooring was just to make the people working there look taller and more impressive, kinda like how they do with pharmacists.
Another big reason for raised floors is to handle wiring. I know companies where it was installed only for this reason. Cooling wasn't even on their minds.
Proverbs 21:19
If we get rid of the raised floors, how am I supposed to impress people with my knowledge of zinc whiskers?
There are two types of people in the world: those who divide people into two types and those who don't.
I interned at ARL inside of Aberdeen Proving Grounds this past summer and when touring the supercomputer room (more like cluster room these days), the guide said they used one of the computers in the room to simulate the airflow in that room so they could align the systems for better cooling. How geeky is that!
It's in lower power chips, more efficient PSUs, and possibly liquid cooling where the radiator is outside the building (or a heat exchanger to heat pump loop in hot climates).
To paraphrase a popular saying: "It's the COMPUTERS, stupid!"
Inefficient architectures must be discarded to make way for more modern, smaller, COOLER processors.
Let's address the real problem here -- not the SYMPTOM of hot air.
We need to address the COMPUTERS.
If you "get" pointers add me as a friend (116)!
I am waiting for the day where someone invents a computer that doesn't need to be cooled or generate excess heat.
Think about the lightbulb....A standard 60-watt incadescent bulb generate lots of heat. A better design is something like the LED bulbs that generate the same amount of lumens, with much less power, and more importantly little to no heat.
Good design can allow these devices to not generate excess heat, hence eliminating the need for the raised floor.
Just have the whole data center submerged in an inert solution like the one made by 3M (fluorinert?), and have the workers wear scuba equipment.
Most. Efficient. Cooling. Evar!
Someone needs to create an air interconnect standard that lets server room designers snap-on cold air supplies onto a standard "air-port" on the box or blade. The port standard would include several sizes to accomodate different airflow needs and distribution form large supply ports to a rack of small ports on servers. A Lego-like portfolio of snap-together port connections, tees, joints, ducts, plenums, etc. would let an IT HVAC guy quickly distribute cold air from a floor, wall or ceiling air supply to a rack of servers.
Two wrongs don't make a right, but three lefts do.
We had an issue where I once worked because we had so many servers the general server room that many different groups used was no longer adequate for our needs, since we were outgrowing our alotted space. Now instead of building us a new server room with the appropriate cooling (which presumably would have included raised flooring) we got a closet in a new building. This is obviously not much fun for the poor people who worked outside the closet, because the servers made a good deal of noise and even with the door closed were quite distracting.
Now, we had to get building systems to maximize the air flow from the AC vent in the room to ensure maximum cooling and the temperature on the thermostat was set to the minimum (about 65 F I believe). One day, while trying to do some routine upgrades to the server, I noticed things not going so well. So I logged off the remote connection and made my way to the server room.
What do I find when I get there? The room temperature is approximately 95 F (the outside room was a normal 72) and the servers are burning up. I check the system logs and guess what, it has been like this four nearly 12 hrs (since sometime in the middle of the night). To make this worse our system administrator was at home for vacation around X-Mas, so of course all sorts of hell was busting loose.
We wound up getting the room down after the people from building systems managed to get us more AC cooling in the room; however, the point is it was never really enough. Even on a good day it was anywhere from 75 F to 80 F in the room and with nearly a full rack and another one to be moved in there is was never going to be enough. This is what happens though when administrations have apathy when it comes to IT and the needs of the computer systems, particularly servers. Maybe we should bolt servers down and stick them in giant wind tunnels or something...
"Some days you just can't get rid of a bomb."
If something is airtight, putting air in one end will move air out the other end.
The problem lies with larger datacenter environments. Imagine a room the size of a football field. Along the walls are rows of air conditioners that blow cold air underneath the raised floor. Put a cabinet in the middle of the room and replace the tiles around it with perforated ones and you get a lot of cooling for that cabinet. Now start adding more rows & rows of cabinets along with perforated tiles in front of each of them. Eventually you get to a point where very little cold air makes it to those servers in the middle of the room because it's flowing up through other vents before it can get there. What's the solution? Removing servers in the middle of hotspots & adding more AC? Adding ducting under the floor to direct more air to those hotspots? Not very cheap & effective approaches...
We have been using raised flooring in our data center for decades and never had any cooling issues. Granted we have 4 large air handlers for the room but when running a raised floor one must have the proper system in place. Some hardware is designed to get it's air right from the floor and some is not. Our large server racks don't have floor openings so we have vent tiles in the floor on the front side and the servers in turn suck the cool air through. Raised floor is a great place to route cables/power/phones you name it. Just make sure your your air handlers are top notch (audible alarms/water detection/humidity & Temp control).
Not an expert, but I had some HVAC work done recently in my home.
The blower moving the air only has a certain amount of power. Hook it up to a duct ten feet long, and output basically equals input. Hook it up to a duct ten *miles* long -- even a perfectly airtight one -- the power you put into one end will be lost by the other end, because the air molecules lose momentum (and gain heat) as they bounce off each other and the walls of the duct.
Every time a duct turns a right angle, the molecules lose a lot of energy as they largely slam face-on into the duct work. Rounded corners improve the situation, but not perfectly so.
My HVAC designer said that as a rule of thumb, every right-angle turn in a conventional house duct was the equivalent of adding 10 linear feet, in terms of energy lost to heat.
I'm a bloodsucking fiend! Look at my outfit!
The longer the ductwork, the more turns, and the more severe those turns, the more your fans have to work to achieve the same pressure and airflow. This, because of the increased friction in the pipe.
Now admittedly, friction isn't as important to gasses as it is to other states of matter, but it can have an effect, especially in high flow cooling.
Heat rises, our original designs back in 2002 for our data center called for overhead cooling using a new gel based radiator system. It would have been a great solution and caused us to go with a lower raised floor, just for cables and bracing. At the time the cost was too extreme to justify the design so we went back to traditional raised floor.
Tile placement on a raised floor is key, only allowing the cool air to be pushed up in the front of your racks and creating hot rows facing your exhaust ends into the same isle. This way the cool air is pushed up from the floor, pulled in through the rack by the server fans, and exhausted, where it can then rise to a vent.
To answer the original question, I think that using raised floors for cooling is not the most efficient solution. Top down chillers that address the heat that is rising off the servers would be better. I just don't know that the price of these solutions has reached a balance for the savings. Even with this design you need something to create a cool pad for your racks to sit on. Many times this can just be the concrete slab of the floor.
True, but IMO not the best way to handle wiring, overhead runs are much easier and cleaner. Every raised floor environment I have worked in was a mess under the floor and a nightmare to run new cables through.
If cooling is not a concern, concrete slab with overhead runs is the best way. If cooling is an issue, use raised floor, for cooling only and overhead runs for cables.
Where I've worked it was primarily for running wires, not cooling. I've also worked in places that have the overhead baskets, and quite frankly, although they are convenient, they are 'tugly. They are great for temporary installations and where stuff gets moved alot, but I'd rather have my critical wires away from places where they can get fiddled with by bored individuals.
So, no, I don't think they will be obselete any time soon. But hey, I'm an old punchcard guy.
"Some say it isn't practical to expect air to make several ninety-degree turns and actually get to where it's supposed to go."
I wonder how all those ducts throughtout America (with tons of 90 degree turns) carry air that heats and cools houses and office buildings every day?
I'm in a data center right now with two rack mounted clusters and three IBM Z series machines plus a load of other kit. Without the raised flooring AND the ventilation systems things would get pretty toasty here but it has to be done right. The clusters are mounted in back to back Compaq network racks which draw air in the front and push it out the back. We therefore have 'cold' isles where the air is fed in through the raised floor and 'hot' isles where the hot air is taken away to help heat the rest of the building.
The only other option would be water cooling but that's viewed by my bosses as supercomputer territory.
Ed Almos
The more corrupt the state, the more numerous the laws. - Tacitus, 56-120 A.D.
Very difficult to track down random machine failures to bad interior decoration choices!
An Indian-American Hindu committed to non-violent thought/speech/action alarmed by the global explosion of radical Islam
SGI for example has done just that in some configurations.p uter/2009-1039_3-5428431-6.html
Check http://news.com.com/Photos+SGIs+Columbia+supercom
it can turn on a dime, but also stay on that dime. poor circulation results. trumpets have nice (if tight) curves, and even building ducts can have redirects inside the otherwise rectangular ducts to minimize trapped airflow in corners. for the most part even those corners are curved to help the stream of air.
most server rooms aren't part of the duct, for example, the one here is large and rectangular, with enormous vents at either end. not very well designed.
airflow is a very complicated problem, my old employer had at least three AC engineers on full time staff to work out how to keep the tents cold ( I worked for a circus, hence the nick.) the ducting we had to do in many cases was ridiculous.
why do you think the apple engineering used to use a cray to work out the air passage through the old macs. just dropping air-conditioning into a hot room isn't going to do jack if the airflow isn't properly designed and tuned. air, like many things, doesn't like to turn 90 degrees, it needs to be steered.
-- it's ridiculous how many people misspell ridiculous... (damn, damn, damn...)
Cooling, IMO, is a secondary use of raised floors.
The real usefulness is the ability to run cabling from any point A to any point B in the floor space.
That's good to an extent, as long as the cable runs aren't too long. Go take a look at an enterprise grade colocation hosting facility and you may change your mind. I've spent a lot of time at one of the top-tier MCI facilities. It has a raised floor that's used for cooling and power distribution, but all networking is done via 3 or 4 layers of overhead cable trays. It's much easier to climb on top of a cable ladder that can easily support your weight to run a cable the length of a datacenter than it is to crawl underneath a floor trying to fish a cable past supports, power lines, etc.
We worked very closely with Liebert ( http://www.liebert.com/ ) when we recently rennovated our data center for a major project. The traditional CRAC (Computer Room AC) units supplying air through a raised floor is no longer viable for the modern data center. CRAC units are now used as supplemental cooling, and primarily for humidity control. When you have 1024 1U, dual processor servers producing 320 kW of heat in 1000 sq ft of space, an 18 inch raised floor (with all kinds of crap under it) is not adequate to supply the volume of air needed to cool that much heat in so small a space.
We had intended to use the raised floor to supply air, but Liebert's design analysis gave us a clear indication of why that wasn't going to work. We needed to generate air velocities in excess of 35 MPH under the floor. There were hotspots in the room where negative pressure was created and the air was actually being sucked into the floor rather than being blown out from it. So, we happened to get lucky as Liebert was literally just rolling off the production line their Extreme Density cooling system. The system uses rack mounted heat exchangers (air to refrigerant), each of which can dissipate 8 - 10 kW of heat, and can be tied to a building's chilled water system, or a compressor that can be mounted outside the building.
This system is extremely efficient as it puts the cooling at the rack, where it is needed most. It's far more efficient than the floor based system, although we still use the floor units to manage the humidity levels in the room. The Liebert system has been a work horse. Our racks are producing between 8 - 9 kW under load and we consistently have temperatures between 80 - 95 F in the hot aisle, and a nice 68 - 70 F in the cold aisles. No major failures in two years (two software related things early on; one bad valve in a rack mounted unit).
But it also eliminates the joy of making fun of coworkers who gets lost in a raised floor, or closing them in when they go on a hunt for something...
(S(SKK)(SKK))(S(SKK)(SKK))
Standard racks tend completely to ignore this. They rely on the internal modules handling their own airflow with fans, which is fine if the inlet area to the modules is much less than the size of the duct entering the cabinet. But if the total area of the inlets to the modules is more than the incoming duct area, the modules furthest from the duct (i.e. the ones at the top) will be starved of air. 1U servers are inevitably going to worsen the problem because they create a large number of competing inlets, stratified up the cabinet. Sucking air out at the top will only work if the air flow is so great it creates a significant pressure drop across the servers, which leads to noise problems, is inefficient, and may adversely affect local cooling inside the server. Blades are potentially much better because, with fewer modules in the cabinet, each with similar requirements, it should be easier to design a cabinet-wide ducting system. However, the most logical solution is to go back to designing the entire cabinet as an integrated system - in which case the entire base of the cabinet can be the inlet duct opening, with appropriate internal structures and blade design to fulfil the objectives of keeping consistent flow to each blade rack and across each blade.
It's the old engineering issue - ad hoc design leads to suboptimal results, and systems need to be considered as a whole. Blades are, depending on how you look at it, a step in the right direction or a return to the way things used to be designed when real computers were loads of tight packed boards full of ECL and proper cooling design of the cabinet was essential if the thing was to work at all.
Pining for the fjords
Oxygenate the fluid, and you can even dispense with the scuba gear...
Going to have to bugdet for towels, though...
One the facilities (gov't) that I have meetings at, has a raised floor covering the entire building. Yes, it is one story and you can place a few football fields in it. They have the ventilation in the overheads and the cabling run under the floor. It works nicely for them, and provides a clean appearance for the entire facility.
http://ask.slashdot.org/article.pl?sid=05/02/23/19 56217&tid=164&tid=4
An Ask Slashdot question was posted back in February about the merits of raising a floor on a budget. Not a dupe, but a complimentary article.
Wiring is now usually ABOVE the equipment, and with 10Gigabit copper, you can't just put all of the cables in a bundle any more, you have to be very careful.
It's a brave new datacenter world. You need some serious engineering these days, guessing just isn't going to do it. Hire the pros, and save your career.
--Mike--
So long as you have positive air pressure under your floor, you'll get *some* effect from your perf tiles. However, as I'm sure some fluid dynamics folks will jump in with, air flow is a HARD problem. Yeah, so you're getting cold air coming up through your perfs. Well, most of them. Some of them are actually pulling air DOWN. Why?
:)
If you're bored, check out TileFlow. It's an underfloor airflow simulator. You put in your AC units, perf tiles, floor height, baffles, you name it. It will (roughly) work out how many CFM of cold air you're going to see on a given tile. It's near-realtime (takes a second to recalculate when you make changes), so you can quickly add/remove things and see the effect. I spent some time messing with this a couple of years ago, and it's very easy to set up a situation where you have areas in your underfloow with *negative* pressure.
The article basically summed it up for me:
McFarlane said raised floors should be at least 18 inches high, and preferably 24 to 30 inches, to hold the necessary cable bundles without impeding the high volumes of air flow. But he also said those levels aren't realistic for buildings that weren't designed with that extra height.
I'd go with 24 inches MINIMUM, myself. Also, proper cable placement (ie: not just willy-nilly) goes a long way towards helping airflow issues. Like they said though, you don't always have the space.
Of course, with the introduction of a blade chassis or 4, you suddenly need one HELL of a lot more AC
Endless arguments over trivial contradictions in books written by ignorant savages to explain thunder in the dark.
When you are designing for a space (such as a room) you design for the shortest amount of ductwork for the greatest amount of distribution. Look up in the ceiling of an office complex sometime and count the number of supply and return diffusers that work to keep your air in reasonable shape. All of the ducts that supply this air are smooth, straight and designed for a minimal amount of losses.
All air flow is predicated on two imporant points within a given pipe (splits and branching with in the duct work is not covered here): pressure loss within the pipe and how much power you have to move the air. The higher the pressure losses, the more power you need to move the same amount of air. Every corner, turn, rough pipe, longer pipe all contribute to the amount of power needed to push the air through at the rate you need.
Where am I going with all of this? Well under floor/raised floor systems do not have alot of space under them and it is assumed that the entire space under it is flexible and can be used (ie no impediments or blockages). Ductwork is immobile and does not appreciate being banged around. Most big servers need immense amounts of cooling. A 10"x10" duct is good for roughly 200 CFM of air. That much air is good for 2-3 people (this is rough, since I do not have my HVAC cookbook in front of me.. yes that is what it is called). Servers need large volumes of air and if that ductwork is put under the floor, pray you don't need any cables in that area of the room. Before you ask: Well why don't we just pump the air into the space under the floor and it will get there? Air is like water, it leaves through the easiest method possible. Place a glass on the table and pour water on the table and see if any of the water ends up in the glass. Good chance it ends up spread out on the floor where it was easiest to leak out. Unless air is specifically ducted to exatcly where you want it, it will go anywhere it can (always to the easiest exit).
Ductwork is a very space consuming item. Main trunks for 2 and three story buildings can be on the order of four to five feet wide and three to four feet high. A server room by itself can require the same amount of cooling as the rest of the floor it is on. (ignoring wet bulb/dry bulb issues, humidity generation and filtering, we are just talking about number of BTUs generated). A good size server room could easily require a seperate trunk line and return to prevent the spreading of heated air throughout the building (some places do actually duct the warm air into the rest of the building during the winter). Allowing this air to return into the common plenum return will place an additional load on the rest of the buildings AC system. Place the server on a seperate HVAC system to prevent overloading the rest of the building's AC system (which is designed on a per square foot basis assuming for a given number of people/computers/lights per square foot if the floor plan does not include a desk plan layout).
Architectural plans are like computer source code with a couple of differences: You only compile once.
That only works until you have a situation where you need to cut the green wire with the yellow stripe, NOT the black wire with the white stripe, in order to shut down your server before it explodes. That oxygenated fluid is pink, making colour detection damn near impossible.
Now, if you're willing to host an alien spaceship at the bottom of your datacentre, maybe they could lend a hand...
Endless arguments over trivial contradictions in books written by ignorant savages to explain thunder in the dark.
Raised flooring is useful for several reasons, moving cool air through a data center is only one of them. While requiring air to make severe turns to get out of the floor isn't optimal, most cabinets and the equipment in those cabinets is engineered with this in mind. Air is generally drawn in through the front of the cabinet and device and warm air blows out the back. Fans in the equipment pull the air in - the air doesn't have to "turn" on its own again (not that is really did in the first place). Warm air then rises after leaving the device where it is normally drawn back into the top of the AC unit.
Raised flooring also provides significant storage for those large eletrical "whips" where 30A (in most US DCs any how) circuits are terminated as well as a place to hide miles of copper and fiber cable (preferably not too close to the electrical whips). Where else would you put this stuff? With high density switches and servers, we certainly aren't seeing less cable needed in the data centers. Cabinets that used to hold five or six servers now hold 40 or more. Each of these needs power (typically redundant) and network connectivity (again, typically redundant), so we actually have more cables to hide than ever before.
Cabinets are built with raised flooring in mind. Manufactureres expect your cabling will probably feed up through the floor into the bottom of the cabinet. Sure, there is some space in the top of the cabinets, but nothing like the wide open bottom!
Anyhow, there you have the ideas of someone who is quickly becoming a dinosaur (again) in the industry.
I don't believe that there should be a rats nest of cabling _anywhere_ in a datacenter. I hate raised floors because they allow techs to get sloppy. Vertical wiring trays eliminate that possibility by showing their hackish wiring job to everyone.
When your datacenter is new, you should pre-wire patch panels in each cabinet for SAN and Ethernet. Each cabinet should have a PDU.
Run all of the cables from all of the patch panels back to your main SAN and Network patch panels.
If you do that work ahead of time, all you will ever have to do is plug a server into a patch panel in the same cabinet.
For larger equipment (Disk arrays, Tape Libraries, etc) you place the equipment and carefully measure the cable runs. Make sure you only have 3 feet of 'slack' and run the cables cleanly.
Its a lot of work to keep a datacenter in order, but it is worth it in the long run. For one, you'll never have to spend two weeks tracing an ethernet cable around the datacenter to locate a phantom server.
Efficiency
LEDs are certainly better than flashlight bulbs.
But when a white LED delivers 15-19 lumens per watt, its about the same as a 100W incandescent and five times worse than a fluorescent. LEDs appear bright because they put out a fairly focused beam - not because they put out lots of light.
Hey, McCloud, get offa my ewe!
77 HITS
Really Long Off Topic Combo
i'm more concerned about keeping my booze cool than hiding bodies. the bodies can be dissolved in caustic soda and flushed down the toilet
What ? Me, worry ?
I wouldn't use water but something that if a leak occurs nothing bad happens. Anti-freeze is pretty much inert and transfers heat well. IIRC, some of the Cray supercomputers were water cooled. So I guess that technology belongs to SGI (for now) since they bought Cray.
Another big reason for raised floors is to handle wiring.
or pluming. I'm serious. (An a bit OT)
When I was at IBM's Cottle Rd. facility, now (mostly) part of Hitachi, they had just finished rebuilding their main magnetoresitive head cleanroom (Taurus). They took the idea from the server techs, and dug out eight feet from under the existing cleanroom (without tearing down the building) and put in a false floor.
All of the chemicals were stored in tanks under the floor. Pipes ran veritcally, and most spills (unless it was something noxious) wouldn't shut down much of the line. It was a big risk but, if what I hear is correct, people still say it's the best idea they had in a while.
I'd rather have someone respond than be modded up.
There are a number of slashdot visitors that do actually care about server room issues. The fact that you don't understand the need does not negate it's importance.
Large organizations rely on server rooms for their computing environment. Having a cobbled environment where the file server is on the 3rd floor, and the application server is in the janitor's closet, etc. is a recipe for disaster. Troubleshooting connectivity issues (among others) can end up costing more than the apparent simplicity of such a design.
Understanding ways to better cool the space that our servers occupy is important. And being able to do so in a cost effective manner is also important. The organization that I work in has one in-house server room (containing 60 racks of servers), and one 'co-located' server room (containing 72 racks of servers). Heat and power are the two killers. If we experience a 50% power loss (assume that one power grid is knocked out), do we have enough power to run AND cool the server room? If not, what percentage of my gear do I need to shut down in order to prevent overheating, without impacting critical business systems (like payroll).
If we can find a cheaper / better / more cost effective method for cooling that utilizes less power, or find a way to use the cooling systems that we have in a more efficient manner, is that not worth an article on slashdot?
IMHO, This is a valid topic.
Raised floor cooling was designed back when the computer room held mainframe and telephone switch equipment with vertical boards in 5-7 foot tall cabinets. The tile was holed or removed directly under each cabinet, so cool air flowed up, past the boards and out through the top of the cabinet. It then wandered its way across the ceiling to the air conditioners' intakes and the cycle repeated.
Telecom switching equipment still uses vertically mounted boards for the most part and still expects to intake air from the bottom and exhaust it out the top. Have any AT&T/Lucent/Avaya equipment in your computer room? Go look.
Now look at your rack mount computer case. Doesn't matter which one. Does it suck air in at the bottom and exhaust it out at the top? No. No, it doesn't. Most suck air in the front and exhaust it out the back. Some suck it in one side and exhaust it out the other. The bottom is a solid slab of metal which obstructs 100% of any airflow directed at it.
Gee, how's that going to work?
Well, the answer is: with some hacks. Now the holed tiles are in front of the cabinet instead of under it. But wait, that basically defeats the purpose of using the raised floor to move air in the first place. Worse, that mild draft of cold air competes with the rampaging hot air blown out of the next row of cabinets. So, for the most part your machines get to suck someone elses hot air!
So what's the solution? A hot aisle / cold aisle approach. Duct cold air overhead to the even-numbered aisles. Have the front of the machines face that cold aisle in the cabinets to either side. Duct the hot air back from the odd-numbered aisles to the air conditioners. Doesn't matter that the hot aisles are 10-15 degrees hotter than the cold aisles because air from the hot aisles doesn't enter the machines.
Moderating "-1, Disagree" is simple censorship. Have the guts to post your opinion.
I wouldn't use water but something that if a leak occurs nothing bad happens. Anti-freeze is pretty much inert and transfers heat well.
Water (non-pure... which it will be as soon as it hits your computer) conducts electricity.
Antifreeze is not better and conducts electricity.
The liquid you're looking for is fluorinert, but the price is one the order of hundreds of dollars per gallon.
When you consider the price, you'll see why many people just use water and high-quality plumbing. Why use $500 of flourinert to protect a $500 computer? If your plumbing fails it's still going to overheat, so the tradeoff is really between the cost of the computer times the likelihood of the pipes failing vs the cost of the flourinert and the likelihood of the pipes not failing.
Since your pipes are more likely to not fail than fail (unless you just totally suck at life), and the flourinert is going to cost more than your computer, it just doesn't make sense.
Life is too short to proofread.
A place to hide the bosses body and won't stink up the place :P
Note that I'm not calling the parent poster stoopid, but rather the design of forcing cold air through the *floor*. As the parent here notes, cold air falls. This is presumably why most home fridges have the freezer on top.
I was most surprised to read this article. I've never worked in a data center, but I have worked in semiconductor production cleanrooms, and given the photos I've seen of data centers with the grated flooring, I guess I always assumed the ventilation was handled the same way as in a cleanroom -- new air in from the ceiling, old air whisked away through the floor. (This ensures that any particles, which will naturally fall if heavier than air, will be sucked out of the room.) Note that this is obviously *not* a passive system designed to use convection, but rather an active system using lots of fans.
While a passive convection system with the cold pulled up from below is a nice theory, you can run into the same problems others have pointed out -- what if the bottom units suck in all the cold air? The top units are left too warm.
Meanwhile, if you drop cold air from above, sure, the top units might suck a lot of that in -- but any cold air that isn't sucked in will naturally continue to drop relative to warmer air, ensuring that the lower units are not cooked. If you want to be especially careful about it, you could route all the cold air outputs towards the perimeter of the room and put the uptakes in the center of the ceiling to ensure a vortical flow.
Just my ¥2.
"What in the name of Fats Waller is that?"
"A four-foot prune."
I spent the first 8 years of my professional life stuck working in NOCs with standard raised flooring, the cooling was just one of the many things it was needed for.
Examples:
Wiring: Not everyone likes to use overhead ladders to carry cables around. In the Army we had less than 50% of our wiring overhead, the rest was routed thru channels underneath the raised flooring.
HVAC Spill protection: Many of our NOCs had huge AC units above the tile level, and these things could leak at any moment. With raised flooring the water will pool at the bottom instead of run over the tiles and cause an accident. We had water sensors installed, so we knew we had a problem as soon as the first drop hit the floor.
If the natural airflow patterns are not enough for a specific piece of equipment, it does not take a lot to build conducts to guarantee cold air delivery underneath a specific rack unit.
The one thing I did not like about the raised floors was when some dumbass moron (who did NOT work within a NOC) decided to replace our nice, white, easy to buff tiles, with carpeted tiles. 10 years later and I can't still figure out why the hell would he approve that switch, since our NOC with its white tiles looked fricking gorgeous just by running a buffer and a clean mop thru it. The tiles with carpeting were gray so they darkened our pristine NOC.
I bet many of the people against raised flooring are land lords that don't want to get stuck with the cost of rebuilding flooring if the new tenant does not need a NOC area. I have been to a NOC in a conventional office suite, they basically crammed all of their racks into what seemed to be a former cubicle island. The air conditioning units were obviously a last-minute addition and it looked like the smallest spill would immediately short the lose power strips on the first row of racks in front of them. Shoddy as hell.
Pedro
----
The Insomniac Coder
iirc its very very difficult to breath under even a fairly small depth of liquid unless the air is presurised to match the water pressure
note: i'm known as plugwash most places but i screwd up registering that here somehow in the past and now can't register
I don't think it is a very good idea to hide your cabling, either power or data. Raised floor just becomes a place to hide things and collect dust, and makes it much harder to make changes. I've seen shallow raised floor which could not be re-seated after it was pulled because of the volume of cable underneath. I've also seen a raised floor environment that became a hazard when the Loma Prieta earthquake popped up every fifth tile or so.
I believe the idea of hiding cable came from early IBM promotional photos that showed a beautiful sea of white tile with an IBM-logoed monolithic rectangular solid standing there in all of its phallic glory. The purchasers, who were not the operators, came to see this as a natural way to install and manage hardware. In my high school days I saw a Sperry Univac 1107 that was not only mounted on raised floor, but actually had components installed in decorative columns that matched the building deco, kind of like a light switch would be in an office -- the whole room became the computer.
Cabinets also make little sense. Why make it hard to connect, disconnect, mount or dismount your hardware? The telcos have been using open racks since the beginning of time -- a much more efficient way to handle hardware that changes or must be inspected frequently.
Power and data should run in separate ladder/tray overhead, where it can be seen and pulled, inspected or added to easily. 20A or 30A power outlets installed in the tray (or overhead duct dropped from the ceiling where electrical codes require) make it easy to attach your cabinet (or better, relay rack) power distribution.
Furthermore, if you speak to the insiders at most of the modern equipment manufacturers, they will tell you that the benchmarking processes are now done on solid, non-raised floor environments. The assumption is tonage of cooling is provided at the intake, which is not located at the bottom of the larger machines, but at the front or back. The hot aisle/cold aisle methodology is still the only viable means for cooling high power density equipment in a large datacenter environment. The only remaining issue is how to get rid of the hot air, and clearly the simplest initial design criteria should be high ceilings (hard to find in datacenters). Outside of that, high velocity air, specially designed air returns, or compartmentalized racks with dedicated air returns are alternatives. In most flow dynamic studies, you find raised floors are riddled with statification, hot air being delivered back into the intakes of other gear, whereas in non-raised hot-aisle/cold-aisle, this problem magically goes away...
Well, you're close. You are correct that the answer lies in a "hot aisle/cold aisle" configuration. The difference is, it works better when the cold air is coming up from below the raised floor tiles.
Why? You must keep in mind, you're not trying to pump "cold" air in, you're trying to take heat out, and as Mother Nature knows, heat rises. So why not harness the natural convection of heat, allow it to flow up to the ceiling, and have some "perf" ceiling tiles and use the space over the ceiling tiles as your return plenum. Thus, you end up with a positive pressure beneath the raised floor, your heat load in the data room, and your negative pressure over the ceiling tiles leading back to your CRAC units.
I assure you it works fabulously in our 2 year old data center at a major financial company. The other advantage to raised floor is, you don't have to worry about water being overhead. No one wants a condensor water, chilled water, or glycol pipe bursting over a row of server racks. But, put your power whips in liquid-tite conduit, use cable racks for the CAT5 and there's no problem if you have a leak. The leak detection ropes pick it up and you can contain it before it becomes a problem.
TileFlow is excellent for most conventional datacenters and has served us well. Best practice in a conventional datacenter is to use TileFlow to ensure positive pressure and appropriate distribution (don't forget to model all your underfloor blockages), set up cold rows and hot rows, and then use a drop ceiling as a plenum with strategically located grates to pull the hot air back into your CRAC units. The only problem, as others have pointed out, is that it's only a 2D model presently and you really need to start modelling your load in 3D when you get past around 12kW/rack.
We've speculated that if we were to start all over again we'd skip the raised floor and do a bi-level drop ceiling with one level being cold air distribution to cold rows and the other level being the hot air return. Let cold air fall and warm air rise, and augment it all with XDO's from Liebert.
that'd apply if you're forcing air into a pipe and watching it come out of the other end. the issue is that if you're forcing air into, say, an underfloor system that's full of different shapes, air passages, etc that the airflow will tend to the path of least resistance, and you'll get less airflow in pockets, which might cause problems.