Janet Jackson had the power to crash laptop computers
A colleague of mine shared a story from Windows XP product support. A major computer manufacturer discovered that playing the music video for Janet Jackson’s “Rhythm Nation” would crash certain models of laptops. I would not have wanted to be in the laboratory that they must have set up to investigate this problem. Not an artistic judgement.
One discovery during the investigation is that playing the music video also crashed some of their competitors’ laptops.
And then they discovered something extremely weird: Playing the music video on one laptop caused a laptop sitting nearby to crash, even though that other laptop wasn’t playing the video!
What’s going on?
It turns out that the song contained one of the natural resonant frequencies for the model of 5400 rpm laptop hard drives that they and other manufacturers used.
The manufacturer worked around the problem by adding a custom filter in the audio pipeline that detected and removed the offending frequencies during audio playback.
And I’m sure they put a digital version of a “Do not remove” sticker on that audio filter. (Though I’m worried that in the many years since the workaround was added, nobody remembers why it’s there. Hopefully, their laptops are not still carrying this audio filter to protect against damage to a model of hard drive they are no longer using.)
And of course, no story about natural resonant frequencies can pass without a reference to the collapse of the Tacoma Narrows Bridge in 1940.¹
Related: Shouting in the Datacenter.
Bonus chatter: Video version of this story and a Twitter poll.
Also, Larry Osterman had a similar experience with a specific game that crashed a prototype PC.
Follow-up: Janet Jackson had the power to crash laptop computers, follow-up.
¹ Follow-up 2: Yes, I know that the Tacoma Narrows Bridge collapse was not the result of resonance, but I felt I had to drop the reference to forestall the “You forgot to mention the Tacoma Narrows Bridge!” comments.
Years later, in 2008-09, the very first 9″ netbooks from a well known brand suffered the same problem. Well, not all of them. The SSD-based model 110 had no trouble at all, but the hard drive-based 150 crashed or even silently corrupted drive data if you played music (almost any music) at high volume. The problem was that the computer was initially designed around an SSD and then adapted for a hard drive (they even had to add a “hump” to the underside so they could fit a standard 2.5″ drive). That meant the right speaker ended just next to the hard drive, which was affected by all kinds of vibrations when playing loud sounds. The manufacturer never acknowledged the problem, but, fortunately, if you knew it, solution was simple: just setting the balance all the way to the left, deactivating the offending speaker. A better and more permanent solution was to replace the hard drive with a SATA SSD, which had the added benefits of reduced power consumption and hit resistance.
That problem, plus the unreliability of sleep/wake with early BIOS versions, gave those models a bad reputation. After upgrading the BIOS and installing an SSD (or deactivating the right speaker) those machines were rock solid (I know because I have a couple of them still working). But the damage was already done.
It’s a widespread misunderstanding that “Galloping Gertie” fell from simple mechanical resonance. To quote Wikipedia (see footnote), “In many physics textbooks, the event is presented as an example of elementary forced mechanical resonance, but it was more complicated in reality; the bridge collapsed because moderate winds produced aeroelastic flutter that was self-exciting and unbounded: For any constant sustained wind speed above about 35 mph (56 km/h), the amplitude of the (torsional) flutter oscillation would continuously increase, with a negative damping factor, i.e., a reinforcing effect, opposite to damping.”
(Also years ago I got to drive over the replacement bridge over the Tacoma Narrows, and my husband couldn’t understand why I was so excited…)
Janet Jackson had the power to crash web servers too. Around the turn of the century, I was part of the team that managed her website. Between her, Eminem and maybe a couple others under the Interscope label, the amount of traffic they generated was mind blowing. Each artist on a busy day could generate GBs of IIS logs, per server. We started on four load balanced NT4 IIS servers and that was a constant nightmare that seemed to never end. We wrote numerous scripts to keep restarting IIS as it was so unreliable. Even if it wasn’t crashing, it would slow to a crawl frequently. The scripts were getting more “smart” to deal with all these inadequacies, yet still just a terrible solution. We had people dedicated to watching their uptime. Even still, their devs were always in our face about it constantly. As more people came into the cable modem and DSL realm, the issue became more apparent.
The unix admins thought they had a better solution on Linux (Redhat 6.9 IIRC), and it wasn’t 48 hours later before their devs were begging to switch back. That was more about compatibility with the cutting edge development than it was about traffic though. The devs were doing amazing things for the time and dependent on dynamic frameworks like ASP and Cold Fusion. They were doing a lot of Flash too. I remember Janet Jackson’s website would start playing her music when you went to it, and then there was a mixer that you could mix in or out the vocals, drums, and other instruments. I’ll never forget it. Our unix admins were thinking they had an equivalent to those frameworks (which I can’t remember the name now, maybe Chilisoft?).
However (and after consulting with Microsoft), we switched them over to two load balanced Windows 2000 servers (RC, before Windows 2000 was officially released) and the rest is history. Other than keeping an eye on the disk space, we never really had any problems after that. Once we got it locked down (if you remember, everything was wide open on IIS5), Windows 2000 was a real game changer for us. Two IIS5 servers could handle effortlessly what four NT4 servers really struggled with.
In the same ballpark. I recently looked at an interesting case where a desktop alert from a security product triggered audiodg.exe to start (when not already started) in order to play the desktop alert sound. This original alert triggered a memory scan of the all processes running which now included audiodg.exe. This process memory scan triggered the reading of the memory of the audiodg.exe process from user mode, as it’s no longer protected since Win7 with OpenProcess/ReadProcessMemory is fine but would lockup the computer every time but not on all computers.
In this frozen state I was not able to force the bugcheck with the crash ctrl scroll or power button. Even remote kernel debugging with COM/NET wasn’t able to crash the computer when hung. Forcing a full dump of the audiodg.exe process from Task Manager would cause the same issue on the computer. This was a vanilla OS install from ISO of Win 10. Normally when creating a dump of Audiodg.exe with audio playing, you might get a stutter for a second as the process is suspended but in this case on this hardware, hang!
When stepping through the memory addresses being read from a test app (kernel debugging) to read the memory of audiodg.exe, to see the virtual to physical (!vtop), the problem memory address was in the range of the “High Definition Audio Controller” as shown in msinfo32. Suspicion being the hardware has dual port RAM where arbitrary reads can cause issues such as this. This only happened on certain integrated sounds cards as adding a USB sound card was fine. HP Compaq 6000 Pro SFF is one example if you want to test the issue, just try reading all the memory of audiodg.exe, e.g. TaskManager – Create dump it will hang the computer.
Possibly the most intriguing case I’ve looked at.
Long ago at a certain minicomputer maker with HQ in an old woolen mill they noticed that the voice coil disk drives would sometimes get read errors, several of them at once.
After investigating they found that only the drives mounted north-south would fail. Elaborate measures were planned to shield them from magnetic fields.
And then it happened while someone was in the lab, and they found that there was an audible thump at the same time. It turns out that trucks backing into the loading dock only shook the building in a north-south direction.
The drive voice coil issue was worse than that. Cause it was tricky to position drives in that building properly (lots of rooms making orientation confusing), field service folk took to varying a compass so that in that building they could position drives properly.
Problem was, that the “in that building” part got lost along the way and techs were rearranging customer computer rooms to enforce this “rule”.
Cables were big and carefully routed so drive orientation affected everything else. Some customers were upset that their beautiful “corporate jewelry” computer no longer faced the glass window like they planned. That caused folks to root cause the “why does it have to be this way” folklore and ultimately rearrange the customers computer room like originally intended.
I wonder if feng-shui came about because of some long-forgotten case like this. 🤔
It is the Contracrostipunctus all over again!
This blog post became the basis for a whole news story on Ars Technica today:
Old laptop hard drives will allegedly crash when exposed to Janet Jackson music
At least this time Raymond is described as a Software Engineer and not Microsoft Executive.
DailyTech identified me as a Microsoft executive
Mr. Chen, could you please clarify if your colleague in Windows XP product support witnessed this or if this is one of those urban legends that happened to “a friend of a friend”. Also, I’m wondering how you feel about CVE 2022-38392 which is entirely based on this blog entry. While it is hilarious to think that Janet Jackson’s Rhythm Nation is a Denial of Service attack, it seems almost preposterous that a laptop could be that shoddily built. Did they not have proper damping of vibrations?
That said, I did analyze the frequency spectrum using Audacity and the most prominent peak is at 84⅛ Hz, which is a smidge higher than E2 on a piano. As it happens, 5400 Hz is a smidge higher than E6 (you’ll need an extended 108-key keyboard for that one). Being the same note, six octaves apart, I have to admit that if any song is going to cause problems due to resonant frequencies, Rhythm Nation would be it.
This should be easy to test with a signal generator. If one has sox installed, you could simply run
and see what hard drives start logging ATA retries. Or, to be more accurate, one could use
which is 5400÷2⁶, while E2 is 82.40689 Hz.
My colleague claims to have been part of the team that investigated the problem. I accept his word for it. (I also think the CVE is somebody playing a joke.)
The story seems to centre around the video, which contains around 20 seconds of just above sub-bass at around 50Hz to 100Hz with a couple of strong 82.4Hz amongst it. Surprised it didn’t happen more (if the drives were built to not dampen this)!
Thanks for asking this question, this sounds like one of those stories that are too good to be true. If confirmed, it would make an excellent example while teaching the physics of resonances. So any sort of further detail would be very much appreciated.
> Did they not have proper damping of vibrations?
Many consumer laptops even as recently as 2013 didn’t bother with any sort of dampening because that would increase cost. Netbooks (what anecdotes from other commenters seem to indicate was the likely place this occurred) would definitely not have had any due to both size and cost reasons. 10¢ of rubber dampening matters a lot when competing in the sub $300 category sadly. When you build to a price you get what you pay for honestly.
> As it happens, 5400 Hz is a smidge higher than E6.
@microsoftonline-com : 5400 rpm is 90 Hz. How is 5400 Hz relevant? 🙂
Correct me if I’m wrong, but if I recall my physics lessons correctly, resonance has little to do with the speed of rotation of an object**, and more about the natural frequency of vibration for that object, which is linked to (1) its length, and (2) the speed of a wave traveling through that object (the latter affected by object composition, e.g. aluminum, steel, etc.; its stiffness/thickness, i.e. the stiffer it is the faster it’ll travel; and its mass, i.e. the heavier the slower the traveling wave will be). Actually, the length would be a pretty major factor since it would dictate which frequency an object is “vulnerable” to by resonance; alter its length and the “attack” frequency no longer has any effect. Make any sense?
So, for a laptop HDD, for the length we would start with 2.5″ (so says http://18.104.22.168/ref/hdd/op/mediaSize-c.html) *minus* the diameter of the platter’s machined inner hole, with that measure then divided by 2 (i.e. the platter being a 2D annulus, the width of one of the two “bands” when you cut the annulus in half through its center).
But I wouldn’t have a clue as to the natural frequency of a single 2.5″ HDD platter; either it needs to be determined experimentally, or else there may be tables for common materials (with given density) by their thickness (or engineering software stacks/packages capable of computing them).
So, it’s probably more complicated than 5400RPM = 90 Hz.
[**wouldn’t angular momentum make platters *less* susceptible to vibration perpendicular to their surface, i.e. along the normal?]
I think it’s more complicated than that, even. Resonant modes of a flat disc with a hole in it would be determined by a set of partial differential equations that I don’t even want to think about, but the solution would also depend on how the disc is supported – that is, if you took the platter out of the drive and hung it on a string, it would have different resonance than it does when the entire inner circumference is held fixed inside the drive.
It doesn’t have to be the platter, though – most of the drives I’ve torn apart had the heads on the end of long (relatively speaking) flat strips of spring steel, supported only at the end away from the head, so any resonance in them would have had its maximum amplitude at the head end, right where you really don’t want it.
Original message posted on the 1st of April 1997? 😉
Why did they ruined the audio instead of making proper isolation for hard drives though? Filtering out frequencies from the audio without user’s consent because of crappy hardware is a very ugly workaround. And laptops can still be crashed by malicious actor using other audio devices.
There is also the story of the resonant frequency of chicken skulls from the old Borland Turbo C++ documentation:
True story: 7 Hz is the resonant
frequency of a chicken’s skull cavity.
This was determined empirically in
Australia, where a new factory
generating 7-Hz tones was located too
close to a chicken ranch: When the
factory started up, all the chickens
Yes and a colleague once told me gullible had been removed from the dictionary.
I guess they were lucky in the recording studio that none of their machines had such drives. That would be some support call: “Every time we try and record this song all our machines crash!” I can just imagine that support call to Microsoft.
Dang it! I was gonna play it on my phone at work & shut the office down but it’s been resolved! 🤬😠😡
This kind of issues are still very much present in today’s storage servers. I work for a company who specializes in Acoustic simulation and we help some customers to avoid having resonances create reading / writing problems in hard drives. The source of the noise / vibrations is often the cooling fan as companies are trying to compact more and more hard drives in the same volume, putting more constraint on the power for the fan to ensure thermal performances. You can read more here if interested, it’s a short article explaining how Meta deals with this kind of issues: https://www.mscsoftware.com/sites/default/files/optimising-storage-server-chassis-design-with-aeroacoustic-simulations.pdf