How could a server explode? [closed]

Question

In short: there are a few servers in a small server room. Being cheap general not-really-business-oriented machines, they did not necessarily have good safety features from the beginning, and what's more, they could be arbitrarily modified before being brought to the room and installed there. How could such a server die a sudden and dramatic death, with the perpetrator able to make practically any preparations for the sabotage?

I've heard some kinds of chips can explode somehow, but if this scene sounds absolutely implausible, consider what could be different if the servers were manufactured a decade or two into the future.

---

As much as I'd like to learn if blowing up a computer is in any way possible, I also have two conditions specific to my story:

• there must be no visible clue of intentional sabotage — it must look like a technical problem upon visual inspection (so, no explosives inside the server case)

• the attacker must be able to remotely* start the process at short notice, with at most 3-4 days from the start signal to the final catastrophic event

^{_{* Meaning: from outside the room. E.g., hacking the system, changing voltage or air conditioning of the room. No inside traitor, however.}}

---

I'm writing a story involving an Internet-connected server which at some point explodes. The system is optimized for the compilation and execution of simple programs as well as for various computations, working as a cloud testing environment for a local community of computer scientists. (The reason why anyone would want to make it blow up is, frankly, difficult to explain.)

The tiny server room where it is located can be assumed to be reasonably safe (physically), however security practices are not the priority for the inexperienced sysadmin. Therefore, the chance of initial compromisation is very realistic (i.e., the outside person installing the server may loosen some screws or even infect the system, and the sysadmin would never notice). Also, the never-updated system could be hacked at any time, since the server is connected to the Internet.

Whether it was from personal experience or science fiction films, I thought the processor overheating would do the job. For example, a bug in the system could suddenly cause a massive amount of computations to be performed, generating constant intense load on the processor, which would ultimately cause it to overheat (and explode...?) However, it now seems to me that more problems need to arise to lead to such an extreme outcome. Since nobody around the server room really knows or cares what is going on (as long as the server is up), and the computer scientists never visit the room themselves, the attacker has lots of opportunities.

If an explosion is really improbable, a similar dramatic incident could happen instead, e.g. the spinning fans could slip off or the box could fall to the floor or catch fire, leading to disastrous consequences and possibly destroying nearby objects in the process (which, come to think of it, could be the ones exploding subsequently). As I noted, the person setting up the server is free to prepare it for anything.

Answer 1

Backup battery bank failure

As you've noted, it would be difficult to get computer chips to fail - they simply don't contain anything all that explosive. About the best you can do is an electrical fire resulting in exploding capacitors, but they're more like popcorn than an all-out fireball boom. So, we'll have to focus on the surrounding infrastructure - the most likely candidate being batteries.

Datacenters typically have large banks of backup batteries (Google, for example, has a battery in each of their racks) - classic examples look like this:

The great thing about typical batteries is they are rather good at exploding. A novice datacenter admin might try to mimic Google using cheap batteries with imperfections - making their setup rather vulnerable to a major problem.

Remote triggering a battery failure

It would have to be poorly designed but let's try it anyway! So, the first thing that would be needed is the safety vents on the batteries would have to get blocked up. Maybe they could be blocked up by dust - after all, dust is a common problem in servers anyway. The above image is even pretty dusty.

You'll next need to build up some heat around that battery - this part could kind of be performed remotely. Assuming they used a Google-style design and the battery is in the rack, then we could maybe pull it off by making the CPU get way too hot. Disable the fans and overclock it, then make the server run something seriously intensive. If possible, disable the CPU's thermal safety measures (this is hard, but it's likely that it's optional anyway; dust ignites rather easily). Do this to all the racks to make the heat really climb for the one at the top of the cabinet. If you're lucky, dust around one of the CPU's would catch fire; a short amount of time later, that fire engulfs the battery.

Rather usefully, this scenario doesn't necessarily require any physical access too as it can be explained with a series of novice mistakes.

Summary

A stack of poorly designed, very dusty Google replicas with a cheap built in battery. Maybe even place the battery right next to the CPU to give it the best chance it can get!

Next, it sounds like your datacenter isn't big enough to have either of these, but let's consider them anyway.

Diesel generator fire

Many datacenters also have diesel generators as a backup. A somewhat famous example was during hurricane Sandy in NYC when the guys at Squarespace manually hauled diesel up 18 flights of stairs to keep the servers online. This could be used as a nice source of something classically explosive nearby.

Cooling system failure

A major source of cost and complexity in an average datacenter is the cooling system - large pipes filled with warm water.

So, if one of the pipes became clogged in some way and the datacenter builder didn't bother with all the normal safety measures (such as an expansion tank) then the heat would build in the system potentially taking it to breaking point. Water explosions are wonderfully powerful, but there's no fireball of course - so it depends on the kind of effect you're after!

Fire sprinklers

Rather ironically, a fire sprinkler can cause a fire when electronics are involved. Let's say the perpetrator triggers the fire alarm which in turn rains down water on the rack. That generates sparks, which in turn starts an electrical fire. The fire then proceeds to explode one or all of the batteries using the setup above.

In this scenario, for some physical accuracy, the top server would spark but would be too wet to ignite any dust. One of the lower servers would receive enough of the water (but not too much to drench it) to be the actual ignition source.

The bonus of using a sprinkler is it would be a lot easier to pull off remotely; it wouldn't involve breaking basically every safety measure a CPU has. It's also rather subtle - a novice probably wouldn't bother checking the ceiling when positioning their servers.

Or..

Hit backspace:

Shift is even better:

Answer 2

Your server banks are cooled by massive quantities of a flammable refrigerant which ends up getting ignited by a spark from faulty (or sabotaged) electrical wiring.

Answer 3

Not too long ago I had smoke come out of the vent on my computer. It was due to the power connector on an optical drive. It was continuing to burn the plastic, but would not “catch” to burn without the heat source.

I recall when Li-ion batteries were first appearing, an early adopter (because of the low weight to power ratio) was RC aircraft. Many did in fact end in an impressive fireball.

So, being an early adopter of some new technology might be key; it doesn’t have all the safety features yet, they apply the low level tech themselves rather than find an already engineered solution, and they don't appreciate the dangers. It also has the advantage that you make up something new so people don't complain that X doesn't behave that way, since you use Y which is made up.

A true explosion in the technical sense seems unlikely because the materials are not explosives. Or maybe the people combined things that were not meant to be explosive, like how airship waterproofing is what today we call thermite.

Furthermore, a lot of stuff simply doesn’t burn, and that's also by design. Your people might use different stuff rather than normal, not realizing that it compromises normal fire safety.

Finally, you might have an intentional self-destruct mechanism! Yea, they didn’t know that the cheap ebay drives were surplus from government contractors intended for servers on board military spy planes…

Answer 4

The server / computer has none or very few flammable parts

The small batteries are very small and would be hard to over charge

Can get a capacitor to pop but that is far from an explosion

Memory and CPU is mostly sand and it is not going to explode or catch on fire.

A fan might heat up and have a short electrical fire.

The operating systems protects the hardware in that a regular program cannot control hardware directly. It is call the hardware abstraction layer.

I think the best shot would be to hack the EPROM to shut off a fan and hack the thermal sensor to not report over heat. Even access to the EPROM from a regular program is not easy. Hope the EPROM fails as part of the overheat or they just don't check.

An OpenSource OS like Linux you might be able to do it but still you are just going to fry (short) some components. No fire or explosion.

Answer 5

The problem you have with having anything dramatic happen is everything is isolated.

For example, a power supply under the right conditions might catch fire for a bit, but is encased in a metal shell that would limit or prevent its spread. The following is about the best you can get.

Let's pretend the servers were purchased with no name power supplies, and were already running at 105% of max. Due to cheapness, the safe guards provided by normal manufacturers was left out. Now some motherboards allow overclocking via software, so with the correct virus you could potentially push it even further over the top causing the whole system to draw even more power. Then runs software to max the CPU and GPU. You might also be able to A. stop the fans,on the motherboard or B. override any alarms when you force the fans to 1rpm.

At this point you might be able to melt some wires inside the power supply. Largely the CPU have heat sensors built-in, and would automatically throttle the speed down to prevent damage. You would have to be a super genius to alter the CPU as it is all encrypted in multiple layers by Intel.

Maybe you could get the capacitors to pop, and a small fire that would be contained by either the metal case of the power supply or the chassis. An actual explosion would be almost impossible. The super cheap components inside the power supply would quickly melt or short out, and cause an open circuit disrupting the short. Either that or trip the circuit breakers, and that would stop the short.

Typically servers are connected to UPS, which would detect the short and turn off.

If you significantly tampered with the electrical system, maybe you could have a fire, maybe. As in prevent the any breakers in the whole electrical box from tripping.

Your only hope is the building catches fire, and there is a gas station with a giant propane tank outside which the fire can spread to and then EXPLOSION!!!!

Answer 6

Yes, computer chips can explode, but it's a question of scale.

Cooling fans are controlled by the operating system to keep temperatures within specified limits, starting/speeding them up if more cooling is needed, slowing/stopping them to reduce power consumption and noise when not needed. Also, CPUs will slow themselves down if necessary to reduce generation of heat if there is a threat of overheating.

These measures can be subverted; cause the CPU to work very hard, continue to run at its highest speed with all cooling turned off, and the chip will overheat and go into thermal runaway. It will eventually destroy itself explosively. In human terms, it won't look or sound like much, perhaps all the sound and fury of a Christmas cracker.

There may be smoke, and there may be ensuing fire; computers do contain a fair amount of combustibles - plastics in the circuit boards, chip packaging, insulation, and various mechanical parts. Likely not enough to do anything more than release a harmful dose of toxic smoke, but there is potential to start a larger fire if more combustibles are very close by - as in stacks of paper piled all around the machine. For a single node in a server rack, going "poof" like this would probably do nothing to anything else - nothing within range will be combustible.

Answer 7

You'd be surprised how much stuff is handled in software these days. For example, one fellow discovered that a Macbook with a compromised kernel could actually break into the microcontroller in the battery, and reflash its firmware. (The battery was using its own off-the-shelf OS, and Apple hadn't changed the default password on the batteries it shipped.)

In any case, this software is responsible for handling charging and discharging. If I recall correctly, lithium batteries will catch fire if they're discharged too much, and then you try to re-charge them. According to Wikipedia:

Discharging beyond 2 V can also result in capacity fade. The (copper) anode current collector can dissolve into the electrolyte. When charged, copper ions can reduce on the anode as metallic copper. Over time, copper dendrites can form and cause a short in the same manner as lithium.

Answer 8

Besides turning cooling of and overcharging the batteries you could also try to spin the harddrives over their max speed maby, you can get them to shatter. The harddrive shards could also make some big damage allthoug this is very unlikely to happen.