Hacker News Reader

Comment by mschuster91 | Hacker News Reader

All drives broken at the very least, a 10k rpm drive doesn't like ordinary vibrations already, much less 40cm+ displacement in a fraction of a second.

Heavy servers and especially UPS battery rows would likely take the whole rack down/toppling over with them. Plus: most cabling is secured to the walls/floors with not very much slack, which means a LOT of damaged/torn cables and FUBAR'd ports.

All in all, if an earthquake of the scale in the video would hit a typical European DC, it's bye-bye and rebuild.

View on HN

Which gets you into that boring part of actuary calculations. In seismically inactive parts, "bye-bye and rebuild (while the cluster is running from a geographically remote datacenter, and only a slight increase in latency indicates this)" could be an order of magnitude cheaper, and thus preferable. In other words, downtime is not replaceable, hardware is.

Just guessing, but situations like these might be the place where it will make sense first to use server-side SSDs. While probably not there yet, it makes sense as the gap narrows that some day the cost of using SSDs is cheaper than building a whole floating/moving floor.

The SSDs address part of the problem. As noted by another post here, many rooms are build and structured with the assumption that nothing ever moves, so rather reasonably, you lay cable in a manner that has little flexibility, you mount things to floors/walls, and so on.

The data integrity is important, but the physical design and layout makes a rather big assumption that the infrastructure you're mounting stuff on will not move or vibrate.

Plus the DC itself might be running, but now you could be running it on a generator (no outside power); better hope you can get a fuel truck to it before the 1-week tank runs out (are there truckable roads outside now?). And connectivity - you might have line slack inside your building, but the fiber cables might be broken in several locations before they even get to next active element (which is also earthquake-proofed? And powered? For how long?); so perhaps you're down to the radio fallback (hope the antennas aren't misaligned too much, let alone that their masts are still upright); again, assuming that the other end of the link is operable.

In other words, preparing for vis major is far harder than it looks. Doable? Yes, if you're willing to go all the way to military-grade resilient comms...for a military-grade budget. Far easier to failover to a DC in a different zone.

Forgive my uninformed question but I remember seeing a video many years ago where a (Samsung?) hard-drive had built-in protection from being dropped. The disk will stop spinning within a millisecond when it detects movement. They could survive a sudden drop onto a floor. I assumed this was a common feature of modern disk based hard-drives.

Could built-in protections like this in harddrives be a sufficient protection in an earthquake? What other moving parts are that at risk here? Assuming the servers themselves are locked in cages, I wouldn't expect them to crash to the floor.

Although the big difference here is it's not a sudden drop but a prolonged forceful shaking.

That keeps your data safe, but (and I'm guessing here) quite plausibly means filesystem writes may likely start failing, causing services on the affected server to start failing. If you need uptime during an earthquake, then that's not sufficient.

Why guess? From the article on the sensor:

"When the computer is stable, the drive operates normally again."

So unless you install a drive that has its own motion detection on board the computer will simply quickly 'park' the drive and will allow it to resume operation without any issues when it has landed.

to be fair, AlephGarden did say during an earthquake. Telecoms, Hospitals, etc, perhaps wouldn't be happy with being out for the time taken for an earthquake to blow over. Plus aftershocks when damage control is in progress.

They'd be a lot happier with being out for that time rather than having to replace a bunch of hardware.

A typical accelerometer can detect free falling and the hard drive can spin down before it hits the ground. I guess in that sense being shaken is worse than being dropped "suddenly".

Sudden motion detection [1] is built into many notebooks, and also allows you to use your MacBook as a lightsaber.

[1] https://en.wikipedia.org/wiki/Sudden_Motion_Sensor

And thinkpads.

I used it to add X11 autorotation to an X61 tablet. No spin animation, but it was a cool one-day hack nonetheless.

The fun part was that the orientation analyses was done from a shell script, and actually worked 100% of the time (unlike so many phones)! \o/

The insane part was that DRM/KMS on Linux is 1000% crazy and I had to patch the kernel to stop poking the backlight on every. rotation. event, grr. (I had to do it in the kernel - I basically killed DPMS - because X11 libdrm is utterly incomprehensible.)

These were fun, but if i remember rightly, Apple have largely phased the Sudden Motion Sensor out on all their laptop models that come SSD only, which is increasingly almost all of them. There were some ridiculous examples of using it to sort mail into folders in mail.app as well, by tilting the MacBook left and right.

> All drives broken at the very least, a 10k rpm drive doesn't like ordinary vibrations already, much less 40cm+ displacement in a fraction of a second.

Do commercial server drives not have the same anti-shock tech as laptop drives?

Drops are easier because there's a gap of free fall before the actual hit so there's time to sense and park the heads. Earthquakes are more instantaneous where the shaking is the first warning the protection can react to.

Reply on news.ycombinator.com