Earthquakes Shake Servers, Too

Michael Buhrley writes "I felt a pretty good earthquake this afternoon in Tokyo. I immediately went to the Japan Weather Association earthquake information page to see if it had registered the quake, which it had not (the ground was still shaking at this point.) 20 seconds later when I refreshed the page the server had slowed to a crawl. I had been looking at traffic graphs for one of my servers earlier and thought it would be neat to correlate the traffic data with the seismic data for the event. I wonder how quickly a noticeable traffic spike could be detected and what other information could be gleaned from the web behavior. Lots of traffic = big quake or quake in big city. The U.S.G.S. Pasadena Field Office has a page that compares this phenomenon to the Slashdot effect."

Earthquakes and response

[WhiteBandit]

It seems like if you live in a fairly seismically active area, that people are pretty keen on the webpages. The USGS has a site showing earthquakes in California, located here [usgs.gov]. We had a decent sized earthquake here a few weeks ago in Big Bear at about 4:20 in the morning. It woke me up at the end and I couldn't be sure if there was an earthquake or not. So I got up, turned on the computer and sure enough, the site showed the earthquake immediately.

One of the cool things about that site is the fact that you can report what you felt in your area and they create a shake map based on the reports. Within 10 minutes there were already about 15,000 reports that people sent in and that number climbed quite a bit as the morning went on.

Part of my reason for logging onto the site after an earthquake is curiosity. I want to know where it was centered and how big it was. I think that has to do with a lot of other people's reasons for logging on as well.

Re:You must live somewhere besides Tokyo or Calif

[WhiteBandit]

Well technically true :)

The Richter Scale isn't really used anymore. It's originaly purpose was to measure vertical ground motion and it loses its accuracy above something like an 8.5

What is used now is something called the Moment Magnitude Scale that actually computes the amount of energy released in an earthquake. It is fairly similiar to the Richter Scale when you compare magnitudes of earthquakes. Obligatory linkage [ualr.edu]. One thing to note though is that each step up in the MMS is an increase of 30 the amount of energy an earthquake releases. So a 6.0 releases 30 times more energy than a 5.0

Basically what this comes down to is that people will think "Oh good! We had a 6.0 on the San Andreas Fault! So that should release some energy built up!" The SAF is capable of producing about an 8.0 here in Southern California, so it would take 900 6.0's to equal the energy of an 8.0 :)

Slashdot effect

[WatertonMan]

While we call this the slashdot effect lots of other portals do the same thing. For instance when the Elizabeth Smart case happened yesterday most Utah news orginizations had their servers thrashed by over-access. That was in part because Drudge Report (and likely others) linked to local news stories.

As for utilizing this, I suppose you could set up a script that monitors such sites in a manner akin to ping. Although I think that most administrators would prefer you didn't. Get a bunch of such clients going and you effectively have an accidental denial of service attack.

Further such monitoring only works on servers that aren't designed for high traffic. Put an other way, what would cause a slowdown for your local paper is likely very different from what would cause a slowdown to CNN. Further as servers are upgraded you lose your "baseline."

So there is an effect and the effect correleates to what is "significant" to the readers of that site. But doing much with the information would be hard.

Having said that though someone had a joke about an early warning system built by checking Drudge Report (a popular American news portal). It probably is a good idea. When there is breaking news most people go to the Drudge Report because he typically links to the best information about the news. He further tends to put important breaking stories in red.

CDC does similar work

[0x0d0a]

The CDC looks for similar stuff like this -- spikes in reported illnesses, blah, blah, blah.

The way they really have any idea what's going on is massive correllation of many databases. School absenteeism jumps in an area combined with a local increase in NyQuil purchases or something like that.

Re:Not really new...

[oobar]

You can read Taco's account of manning the slashdot servers [slashdot.org] through 9/11. In summary: normal slashdot traffic is about 20 pages/sec and 1.4 million per day. That spiked to about 60-70 /sec on 9/11, with nearly 3 million for the day.

Re:this already happens to the USGS webservers

[stan7826]

I'm the system geek for the USGS Pasadena office, and I wrote the article referenced. I've found that traffic on our servers starts increasing within one minute of the earthquake. The peak traffic for California earthquakes is always ten minutes after the event. You can set your watch by it, it's that dependable.

As for the USGS servers having problems after earthquakes, we've been served through Akamai EdgeSuite since late 2001. So for the most part, our servers have been doing better. We've had a couple of other problems caused by other things, though. I've written reports about a number of these at http://bort.gps.caltech.edu/spikes [caltech.edu]

As for notification, I have one site that is not served by Akamai [ www.trinet.org [trinet.org]] that still uses a Squid server. So I use MRTG to monitor the Squid, and it calls a small Perl script that pages me whenever the traffic on the site increases by more than 10% in one five-minute interval. This always tells me when people have felt an earthquake, sometimes ever before our automatic location systems are able to page us with a location and magnitude.