Slashdot Mirror
Net Traffic Shocks Mimic Earthquakes
belphegor writes "Technology Research News is running an article describing research linking the similarties between Earthquakes and Internet traffic. By pinging hosts across the network, researchers 'were able to measure frequent changes in Internet congestion...results showed that the Internet, like the earth's network of faults, exhibits criticality -- a condition of sudden and drastic change. "Sudden drastic congestion leads to a large value of the round trip time of the ping signal, which is identified with a main shock," said Abe. The researchers referred to these sudden, drastic traffic changes as Internetquakes.' They also saw 'aftershocks' that can be mathematically described in a similar manner to the seismic ones more familiar to many Californians."
This pattern of behavior (where two completelly different things show the same underlying behavior and/or explanation) is exactly one of the things that Stephen Wolfran is trying to explain in his book A New Kind of Science (see amazon link and reviews here)
Basically, everything in the universe can be explained as a huge network of nodes, where all these node do is computations following very simple rules. From such simple rules we get all the laws of physics, human behavior, chaotic behavior, and in this case the behavior of an earthquake and Internet traffic.
I like to think of the net as a non-linear resonant
circuit. The correct starting conditions give
you very large responses to "small" inputs.
It's resonant so you see "ringing" in the response.
Seems to me that you can definitely look at the
earth's crust in a similar manner.
Absolute statements are never true
Now all of the money the government has spent on the detection of earthquales can actually help ME. Advance knowledge of net connection would make my life much easier.
Heck, VOIP might even actually be commercially feasable.
___ I don't respond to Anonymous Cowards, and I Never Mod them UP.
I see the /. effect more like a tornado than an earthquake. An impenetrable wall of traffic is created around the site which sit in the calm "eye" of total disconnection.
...like a lot of work to do to come up with what basically appears to be not much more than a metaphor.
Roving Web-Teleoperated Robot
So these server items Host only web pages, and all the traffic on the internet is web traffic, which moves from link to link by a "router". This is very upsetting that college educations and many many years of research show something that could be acomplished in 20 mintues with a SDSL line, a p0rn site on a connection faster then the SDSL line, and finally root@box'#ping xxx.xxx.xxx.xxx
I wounder if they have their MCSE?
www.oobersworld.com - For those that ride.
Hmmm... I like the idea of an Internet that "rings". But isn't it the "ringing" that we would like to get rid of from the net, in the same way we might wish the earth didn't "ring" so well during earthquakes?
What is the opposite of this model? Something like a marshmallow or ball of silly putty, that deforms to store energy and allows that energy to be released over time? I know that might affect the individual round-trip times, but it might improve efficiency on a larger scale.
I'm not sure what the rules for the nodes would be, but obviously there would have to be a way to balance "slack" and "stress" at different scales simultaneously, in order to avoid criticality. Anyone know about any models that have this kind of behaviour?
I've got a bad attitude and karma to burn. Go ahead. Mod me down.
I always wondered how long it would be before someone tried to give a "weather report" for the Internet. I imagined that some day the daily Internet traffic report would appear right alongside Reporter Bob up in TrafficCopter 7 reporting on the condition of Highway 69 at 8am.
/dev/null. However, I would think it'd be tough to always keep up on what everyone has for capacity unless the backbone providers regularly publish these stats, as well as the stats on current traffic.
"Well, Joan, it looks like Slashdot it at it again. You'll want to keep clear of Alter.net this morning as it appears to be having congestion problems around the Midwest. Communications to anything near a corn field is likely to be slow to smack-me-dead stupid through the 9 o'clock hour. Queue up those emails, folks."
One has to wonder if it's even possible to predict and gauge incoming traffic problems. I guess you would have to know the effective capacity of the Internet, and sub-portions of it, at all times. I can see how a router's effective capacity could be measured by its effective throughput and cache. Your "sentinal level" would occur when the cache is full and bandwidth is maxed and the packets start to get a one-way trip to
I don't know if it would be helpful or not. One one hand, it'd be handy to know that the reason I'm getting 1900ms ping time to SF this morning is because some dumbass tripped on a power cord, but on the other hand, if I really care that much, I can probably figure it out using traceroute et al.
Maybe it's a solution to keep everyone and their dog from flooding the 'net each time a router bites the big one and makes a suburb blink out of existence for a few hours. Other than that, it just sounds like a good excuse to draw pretty graphs.
Blog,Twitter
Slashdot has also been a mitigating factor to some of the most severe 'quakes', especially if you're discussing the information transmitted rather than just network traffic.
Case in point? 9-11-01.
CNN was gone. USA Today was gone. Fox News was gone. Slashdot, however, was providing first-hand accounts of the disaster/attack as well as discussion about them.
The next Slashdot story will be ready soon, but subscribers can beat the rush and slashdot the links early!
In both the internet and the network of earth faults, there are interactions between elements on multiple scales in time and space. That analogy is not hard to guess, but these physicists seems to have contributed a systematic measurement and modeling study. Fine.
So far, just from the reading the press article, my doubts about their work are twofold:
1. their ping frequency is one ping per second. This is a poor sampling of the time series of packet delays. Specifically poor sampling could lead to a false alarm or a miss. False alarm meaning that the sudden drastic 'congestion' could very well be an isolated event suffered by the ping packet or its adjoining ones; the rest of the packet delays in the one-second interval may be small. 'Miss' meaning you didnt catch the congestion due to the poor sampling. The ping packet got through quickly but there were congestion delays suffered by the rest of the packets in its one-second interval.
In short, ping times are a poor indication of the true traffic status of an IP network. One ping per second is especially pathetic if one of the bottleneck links in your route is a high-speed OC-3 or higher link. For example, at 155 Mbps (OC-3), there can be upto 13,000 typical IP data packets through the router in one second. Only one of these is the ping packet!
2. Their flawed analogy between fault energy and congestion level. Fault energy builds up slowly over years and is then released suddenly in an earthquake. They compare to this user sessions running over hours and then suddenly contributing to short periods of drastic congestion. So are they saying that congestion builds up slowly over hours at a router or server as more and more packets are processed? Ha! I am still ROFLing on this one!
There research will vastly improve if they collaborate with internet researchers and engineers who have a working knowledge of the internet's routing and flow control protocols, instead of approaching it from a purely black-box modelling point of view.