Moonshots

Natural Server Cooling

Rob Bell

Issue 2, August 2017

Like many offices, at DIYODE we have a server room. But like all server rooms, it requires cooling - and lots of it! So... is there a better way?

HOT AND COLD

I walked into the office early one morning, and it was freezing cold. It was about 5°C outside, so that’s not really surprising. If you live in a location where you can ski on lakes in winter, this may be balmy weather for you. But here in beach-loving, sun-baking asphelt-melting Sydney, that’s pretty cold to us.

As I walked past the server room to make coffee, I noticed the inside temperature was still sitting at 20°C (and yes, the air conditioner still has to work constantly to keep it there). It got me thinking... why are we using air conditioning to cool the room, when it’s near-arctic outside? In the same morning routine, the air conditioner in the main office gets switched on so everyone isn’t at their computers dressed for the snow. So why can’t we leverage this heat by-product into something useful? Even if we’re using 100% renewable energy to power everything, it still seems like something of a waste.

GETTING HOT IN HERE

The primary issue with server hardware (and virtually all electronics) is heat. As we know, all electronic circuits produce some amount of heat (even if it’s imperceptible in a simple circuit). In most circumstances, the heat is insignificant when considering the room it’s inside, and it can easily dissipate. In an environment like a server room, however, we have an extremely high concentration of electronics in a relatively small space. For that matter, my “very standard” media centre at home (standard cheap graphics card, single Intel i7 processor) quickly shuts down after a few hours of being placed inside an entertainment unit, yet we rarely think too much about keeping a standard computer cool.

Servers often have high velocity, noisy fans to keep the air moving through their critical parts. Facebook is one of the world’s largest data centre operators (not surprising now that they boast over two billion users). They’ve worked hard to keep all those cat videos and photos of your lunch flowing through the internet; but more importantly, have taken great steps to increase efficiency within their data centres. For this reason, they don’t use front panels on their servers, to increase the amount of air that can get through them. Air flow through the server is critical, but if the room starts heating up that air will extract less and less heat from the servers, due to the lack of a temperature differential.

So if it’s all about the air flow, all we need to do is pump the cold air into the server room (or suck the hot air out), and all will be right with the world - yes? Sadly, it’s not quite that simple.

THE COLD HARD REALITY

In theory, with the ability to move an unlimited volume of air rapidly through a server room, you could keep everything within tolerance until the outside temperature is just 1°C below your target temperature, only then requiring air conditioning to take over. But fans use power too... and big fans especially, use a lot of power. There is, therefore, an efficiency point where moving huge volumes of air requires more energy than an airconditioner, which artificially improves the temperature differential.

A quick look at ventilation fans for high volume air transfer, proves how much this is true. We looked at a 400mm unit, which is capable of moving 128m3 of air per minute. That might sound like a fair amount, but the second you add ducting (or even a simple pipe bend for that matter), the volume capability drops rapidly. It also requires a 1500W motor to begin with. If you need a larger fan than this, you’re rapidly matching or surpassing the power consumption of a suitable air conditioner. We’ve also used a few of these large ventiation fans at times, and they are really noisy!

The mathematics required to provide specific answers to this question is a little beyond the scope of our thinking here. Consideration needs to be given to the amount of heat being generated by the servers, the size of the server room itself, the temperature differential between outdoors and the target temperature, the volume of air you can transfer (i.e. the size and capacity of fans extracting the heat), humidity, barometric pressure, friction/pressure incurred in the ducting, and the list goes on. But we’re confident an engineer familiar with air conditioning or thermodynamics could tackle this with little issue.

Now, in experimental scenarios we’ve run our server rack in the main office, which is huge. The sheer volume of air in the office allows the temperature to be maintained at a sensible level. But enterprise-grade server equipment is extremely noisy, so this isn’t a feasible solution either. Sure, you can put noise-reducing insulation around the server rack to reduce the noise, but then we’re back to the original problem of air flow.

This article isn’t designed to deliver a solution; it’s moreso a thought experiment, and will probably forever remain as one. Although, as we know, sometimes brilliant ideas arise from seemingly unresolvable problems...