Too Many Connections Weaken Networks

itwbennett writes "Conventional wisdom holds that more connections make networks more resilient, but a team of mathematicians at UC Davis have found that that is only true up to a point. The team built a model to determine the ideal number of cross-network connections. 'There are some benefits to opening connections to another network. When your network is under stress, the neighboring network can help you out. But in some cases, the neighboring network can be volatile and make your problems worse. There is a trade-off,' said researcher Charles Brummit. 'We are trying to measure this trade-off and find what amount of interdependence among different networks would minimize the risk of large, spreading failures.' Brummitt's team published its work (PDF) in the Proceedings of The National Academies of Science."

Comments

HAHA

[Anonymous Coward]

pnas... lol.

Re:HAHA

[vencs]

Too Many Connections Weaken Immunity

ah-ha!

[binaryhat]

The Goldie Locks network:
"As a first theoretical step, it's very nice work," said Cris Moore, a professor in the computer science department at the University of New Mexico. Moore was not involved in the project. "They found a sweet spot in the middle," between too much connectivity and not enough, he said. "If you have some interconnection between clusters but not too much, then [the clusters] can help each other bear a load, without causing avalanches [of work] sloshing back and forth."

Primitive

[gilgongo]

I'm sure that in 100 years time, people will look back on our understanding of networks, information and culture in the same way as we look back on people's understanding of the body's nervous or endocrine systems 100 before now. This study hints at our lack of knowledge about what the hell is happening.

Re:Primitive

[NoNonAlphaCharsHere]

It would be interesting to apply what they learned here to the power grid.

Re:Primitive

[gl4ss]

the study was about power grids, where it makes a bit more sense. of course in that context(and in data-networks, though with data it actually matters where a certain data packet goes as data consumers don't just want _any_ data, they need specific data, but power you don't much care where it actually came from..).

still, gotta wonder, in real world context you'd need to think about what kind of real mechanisms are used for making the new connections and safeties supposed to stop cascades from spreading.

Re:Primitive

[Anonymous Coward]

Cascades on power networks happen when you suddenly lose source, without rejecting the drain. I.e. the load remains high, but suddenly the flow required to supply that load has to shift because a link went down due to failure/overload.

There is a protection against this, it is called selective load rejection. You shut off large groups of customers, plain and simple. And you do it very very fast. Then you reroute the network to make sure power is going to be able to flow over links that will not overload, and do a staggered reconnect of the load you rejected.

That costs BIG $$$$ (in fines, lost revenue, and indirect damage due to brown-outs), and there is a silent war to try to get someone ELSE to reject load instead of your network. The only thing that balances it are extremely steep fines among the power networks themselves, and in countries that are not nasty jokes, the government regulatory body.

I am not exactly sure how to move that to a BGP4, IS-IS or OSPFv2/v3 network, where instead of a sink pressure, you have a source pressure.

Too many cooks....

[bwohlgemuth]

As a telecom geek, I see many people create these vast, incredibly complex networks that end up being more difficult to troubleshoot and manage because they invoke non-standard designs which fail when people wander in and make mundane changes. And then when these links fail, go down for maintenance....surprise, there's no 100% network availability.

Three simple rules to networks...

Simple enough to explain to your grandmother.
Robust enough to handle an idiot walking in and disconnecting something.
Reasonable enough to be able to be maintained by Tier I staffing.

Re:Too many cooks....

[LordLimecat]

How are you going to explain VLANs, STP, and ACLs to your grandmother? Has it occurred to you that there are, in fact, situations where all of those technologies are useful?

Has it also occurred that if you are properly securing your network with port security, you cant just walk in and plug something in and have it work?

Ooh. I know the answer.

[dgatwood]

Forty-two.

Now we finally know the question.

Other applications of this theory

[dietdew7]

Could these types of models be applied to government or corporate hierarchies? I've often heard about the efficiencies of scale, but my experience with large organizations is that they have too much overhead and inertia. I wonder if mathematicians could could come up with a most efficient organization size and structure.

Re:Other applications of this theory

[Attila Dimedici]

One of the things that technological changes since the mid-70s have taught us is that the most efficient organization size and structure changes as technology changes. There are two things that exist in dynamic and as the relationship between them changes, the efficiency point of organizations changes. One of those factors is speed of communication. As we become able to communicate faster over long distances, the most efficient organization tends towards a more centralized, larger organization. However, as we become able to process information faster and more efficiently, a smaller, more distributed organization becomes more efficient. There are probably other factors that affect this dynamic as well.

Study wasn't about Computer Networks

[fluffy99]

"The study, also available in draft form at ArXiv, primarily studied interlocked power grids but could apply to computer networks and interconnected computer systems as well, the authors note. The work could influence thinking on issues such as how to best deal with DDOS (distributed denial-of-service) attacks, which can take down individual servers and nearby routers, causing traffic to be rerouted to nearby networks. Balancing workloads across multiple cloud computing services could be another area where the work would apply."

The study was about the stability of power systems, which is a completely different animal. For power systems, as demonstrated by a few wide spread outages, are at the mercy of the control systems which can over or under react. Computer networks might have some similarities but trying to draw any firm conclusions from this study would be pure speculation.

I would agree though, that at some point you move beyond providing redundant paths to opening up additional areas of exposure and risk.

Re:Easy if n m bad

[hairyfeet]

Hell ask the P2P guys because if there is anybody that has to balance craploads of connections its those guys. Look at how much overhead the first gen P2Ps used compared to now, with each version they get better at moving data without the connections getting overloaded. Give me somebody that has actually had to deal with the BS day to day than somebody that is writing a paper any day of the week. the trial by fire quickly weeds out the dumb ideas and you fix it or die.