Server rack air flow management

[IanJ]

Hi all,

Just looking for a bit of advice.

I'm building a new electronics workshop (woohoo!).....and as part of this I'm moving my web server's, NAS boxes etc from their current location into the new workshop and into a 32u 19" rack unit. The rack is built and I'm now testing.

My HP 3458a will sit on top of  the rack as I've demo'd in the pic attached. As such the fan vents for the enclosure are at the bottom, and I've mounted a 4 fan 1U unit to pull cold air in from the outside and thrust it upwards. There's a thermometer mounted about 3/4 of the way up the rack. I was hoping that the air flow would push the hot air up and out of the various side vents etc.

Problem is, the fans don't seem to do too much in terms of cooling at the thermometer position anyways, I get better cooling just by opening the server rack door and leaving it open.

The thermometer records about 8degC above ambient, so is at about 29degC, and maybe up to 32degC at it's hottest......which is actually not bad but I'd like it cooler.....especially if and when the PC's start working harder as one is used for 3458a GPIB data logging.

I don't want vents on the top surface of the rack as the 3458a is sat there and don't want it subjected to a hot air flow.

Ideas?

Ian.

[Monkeh]

You could make some openings and add fans to the side or rear panels. Most of the air from those bottom fans will be billowing out the front door and panel gaps.

[IanJ]

Most of the air from those bottom fans will be billowing out the front door and panel gaps.

Interesting!......I wonder if it's worth blocking off the door vents etc so the air flow can only move upwards internally. Easy to test this I guess.

Wish I had a thermal camera!

Ian.

[imidis]

I'd be tempted to remove the door. My rack is smaller but it's in open air.


Either way though, nice rack! 

[BradC]

Racks need to flow front to back. Vertical airflow is always destined to be a disaster. Put a door on the back, put some 120mm or larger fans in the door pulling air through and put a chimney over the fans directing the airflow upwards. Blank unused units and make sure air flows through and over anything that needs it.

[SeanB]

The way that rack is designed it looks like it is meant to have a cold back and a hot front ( or the other way round) with side blanking plates and covers over unused slots. I would probably put the fans at the top, blowing down, and use the bottom exhausts for air outlet. Put filter units on the air inlets to keep duct out, and this should keep the interior dust free. Having air inlets on the carpeted floor will only cause you grief in time, as the fibres will go all through the interior and accumulate all over the equipment interior. Filter the incoming air, blow from the top down and pressurising the case will keep it totally dust and fibre free.

[X]

Try moving the fan tray close to the top and expelling the hot air from the top vents.

Most devices use negative pressure cooling, so the fan will blow outwards, and cool air is sucked in through the vents. You should cool your rack in a similar way, forcing the hot air directly outside, whilst cool air is drawn inwards to drive the pressure system to equilibrium.

Your current setup with the fans drawing air from the bottom is positive pressure cooling, which relies on the hot air being displaced out eventually.

I find the negative pressure method results in fewer hot pockets than the positive pressure method. I think this is because the negative pressure system relies on cool air being sucked in from all over the place, rather than expecting all the trapped air to be displaced by fresh air eventually.

The only advantage I can see with a positive pressure cooling system is easier dust management, since a positive pressure system sucks all the air in through the fan (which can have a dust filter on the inlet) and lets the air blow out of the vents, which is why some industrial equipment uses the positive pressure system.

[CopperCone]

I would not put the 8.5 digit meter ontop of such a noise generator.

1) more noise goes through its emi filter (thats a whole lot of dc/dc converters)
2) possible dc and ac errors depending on  server operation (its a radio) Looks like massive banks of wiring.
2a) hp only put so much money into shielding it. It was intended for high end electronics laboratories of large companies filled with electronics geeks and calibration houses. No one thought it would end up sitting on top of a digital server bank.
3) thermals upset the device
4) dust collection hazard

I would make sure that rack is well sealed, put on a separate distribution line and breaker from your test equipment and kept physically away from the 8.5 digit meter or other lab equipment.

I would recommend accessing it remotely and putting it in another area of your house.

and its a potential hazard because of some kind of RF gasket goes, screw oxidizes, material creeps, etc, iit could start insidiously effecting your measurements. It's not like your going to do periodic studies on the integrity of your rack mount enclosure right.. Plus you got inrush currents into the rack and shit. 

Same goes for the power supplies, one of them could go off and start generating noise due to a multitude of issues.

[IanJ]

Thanks all for the info so far.....certainly some food for thought so I'll continue my testing tomorrow.......plenty of time in hand as it'll be a good couple months before the w/shop is complete and the rack can move in....including the 3458a.

Some additional info. Mains AC testing via my cheap n cheerful power meter......cumilitive readings:

Netgear & KVM = 0.5A / 110W
add 2xQNAP = 0.75A / 170W
add LCD = 0.81A / 170W
add Web server 1 PC = 0.98A / 200W
add Logging PC = 1.2A / 250W
add 3458a DMM = 1.35A / 300W
add 2xfan blower = 1.35A / 300W

So just under 300W total......which is less than I thought it would be.

Ian.

[IanJ]

I would not put the 8.5 digit meter ontop of such a noise generator.

Thats a fair point and definitely something I will put some extra consideration into.......first off I'll run a separate mains AC to the meter rather than plumb it it direct to the rack......and I'll look at an EMI barrier between the 3458a and the top of the rack......and earthing which should suffice.

I had purposely kept the servers towards the bottom half of the rack and just the KVM, keyboard/mouse and document shelf right at the top.

I can't help but think about those photos I've seen of some calib labs where the 3458a etc is just placed in a rack directly next to other instruments...hmmm!

Ian.

[SL4P]

You have equipment installed at floor level, and the rack is installed directly on carpet.
Manage it, or leave the lowest 2RU empty, otherwise you'll be sucking floor and shoe dust into the fans.

[IanJ]

You have equipment installed at floor level, and the rack is installed directly on carpet.
Manage it, or leave the lowest 2RU empty, otherwise you'll be sucking floor and shoe dust into the fans.

There is 20mm gap between the bottom of the rack and the carpet because I raised it on 4 castor wheels.........I also have a couple of filters to install tomorrow so the bottom fans are not sucking up crap......which I am seeing already in just a few days of testing!......but yes, I don't want to be cleaning filters all the time!
The final w/shop won't have a carpet so it should be a bit less of an issue.......but the carpet testing has been a bit of an eye opener for sure!

Ian.

[Gregg]

Brad is absolutely correct about the air flow front to back; you should try to put baffles between the equipment to keep the hot air from the back from recirculating into the front and move the cabinet far enough from the wall to let hot air escape and/or install exhaust fan(s) on the top.
For example most modern data centers are arranged in aisles with all of the equipment fronts toward the cold aisle and all the backs toward the hot aisle.  The best efficiency is attained by closing off the cold aisle as much as possible with doors on both ends and the overhead enclosed; usually the cold air enters the cold aisle from perforated tiles in the raised floor because the total area below the raised floor is the HVAC cold plenum. Great effort is made to seal off the cold air supply from the rest of the room(s) because HVAC is expensive.  In some cases the cabinets have fans on top to help exhaust hot air upwards from the rear of the cabinet which is very helpful with racks totally filled with 1U servers.
A lot of the air movement theory from large data centers can be helpful to even single rack installations like yours.  I’ve seen fan and filter modules designed to push filtered air into the front of a cabinet to aid cooling and minimize having to clean dust out of equipment.  Another concern of blowing air around is zinc whiskers and to a lesser extent tin whiskers.

[Cerebus]

Standard design for rack mounting equipment is to draw cold air in from the front and sides and expel it to the rear. You'll always get some oddball bits of kit where this isn't true, but 95% of kit is designed to this standard.

The most effective cooling method I've encountered is the feeding of refrigerated air into the bottom of the rack from a pressurised false floor. The collocated racks I had in the facility that did this always ran 5-10 degrees cooler (equipment internal temperatures) than the ones I had in facilities that just cooled the room air and left the racks (with fan trays) to deal with things for themselves.

In the absence of luxuries like a cold, pressurised false floor, common practice is to put a fan tray in the 'roof' of the rack and have a 'top hat' vent above it to let the hot air out.

The worst thing you can do is obstruct the air flow in the rear of the rack. That means having enough free space behind equipment to let the air circulate (I'd say six inches or more) and not cluttering this space with cables. Dress cables to the sides of the rack as much as possible. In general, the smoother the air flow, the better.

[floobydust]

It's a really nice rack, just needs to get more air circulating inside/outside.

I see the bottom fan rack as not working, it just recirculates air around the sides.
There is no seal and the fan's inlets need to be piped outside the cabinet with no (easy )air path to the fan's outlet, which causes recirculation.

[IanJ]

Hi all,

Just a thanks for all the input, and an update.

I fitted two 120mm 240V fans on the back panel, near the top. I nice sharp hole cutter made the job easy. Fitted with fan guards front and back.
I have stuck with the original thermostatically controlled fan tray, I just tied in the rear panel fans into the same circuit.
I also constructed a duct in order that the tray fans (horizontal) direct air towards the back of the rack. A cardboard constructed tunnel wrapped in foil tape.

So far so good, with an ambiemt of 24degC I can set the temp at 30degC and when the fans come on it's for no longer than 5 minutes till the temperature is pulled down (1degC hysterysis).
When the room is cooler the fans actually never come on as the temp in the rack floats at just under 30degC....I guess due to the fan holes acting as an exhaust also.

I haven't yet measured to see if the whole unit is now positive or negative pressurized.......I suspect negative pressure as the two rear panel fans are pretty powerful. To help with this I have made some attempt at sealing the bottom holes to the body of the fan tray at the bottom of the enclosure......albeit still need to improve this.

Ian.