Forgot your password?
typodupeerror
Network Networking Science

If a Network Is Broken, Break It More 124

Posted by Soulskill
from the somebody-fetch-my-wire-snips dept.
New submitter Aras Esor writes "When a network is broken — an electrical grid, the World Wide Web, your neurological system — one math model created by a PhD student at Northwestern University suggests that the best way to fix it may be to break it a little more. 'Take the web of interactions within a cell. If you knock out an important gene, you will significantly damage the cell's growth rate. However, it is possible to repair this damage not by replacing the lost gene, which is a very challenging task, but by removing additional genes. The key lies in finding the specific changes that would bring a network from the undesirable state A to the preferred state B. Cornelius's mathematical model (abstract) provides a general method to pinpoint those changes in any network, from the metabolism of a single cell to an entire food web.'"
This discussion has been archived. No new comments can be posted.

If a Network Is Broken, Break It More

Comments Filter:
  • by Roachie (2180772) on Wednesday July 17, 2013 @02:15AM (#44305843)
    ... network break YOU!!!
  • broke (Score:4, Funny)

    by Anonymous Coward on Wednesday July 17, 2013 @02:19AM (#44305855)

    If it ain't broke, don't fix it.
    If it IS broke.... don't fix it.

    • by Anonymous Coward

      If it ain't broken, don't fix it.
      If it can be improved, go for it!

    • Re: (Score:2, Funny)

      by crutchy (1949900)

      If it IS broke.... don't fix it.

      ahh you must work for microsoft

    • If it being broken... grab the popcorn.
    • by Anonymous Coward

      The thing that's broken here, is the analogy.

      It is really easy to see why this is half-thought-through nonsense.
      If you break more, after something is broken, it is by definition guaranteed, that you lose functionality. This is always a bad thing. Again, by definition.

      The logic is like when your car is pulling to the left because of a destroyed wheel, and you then destroy the other wheel too, and declare it a success because now you're "driving" in a straight line again.
      Yeah, but only until you hit the wall

    • by azalin (67640)

      If it ain't broke, don't fix it. If it IS broke.... don't fix it.

      Actually it's more like "If it IS broken, break it more"

    • by OakDragon (885217)
      If it ain't broke ... fix it until it is!
    • by TeknoHog (164938)

      If it isn't broken, don't fix it.

      Fixed that for you.

  • by Anonymous Coward on Wednesday July 17, 2013 @02:34AM (#44305897)
    The plate is chipped. She resents me gluing it back together. So she breaks it real good so I gotta go out and buy an entire new set.
    • by Anonymous Coward
      Non-geek women don't like cracks, wires or antennas.
    • by Anonymous Coward

      Why does everything have to match? Anyways, why don't you buy two identical sets, and that way you have lots of spares.

    • by Darinbob (1142669)

      You know how hard it is to find new Scooby Doo plates? That's why I patch mine.

  • by illestov (945762) on Wednesday July 17, 2013 @02:51AM (#44305933)
    my company's IT department has figured this out a long time ago
  • (Are those genes selfish? Because if so, this) may be a classic case of resolving a Braess's paradox [wikipedia.org] by removing a trigger for selfish behaviour.

    (now, if you'll excuse me, I'll go to RTFA)

    • by c0lo (1497653)
      The first link is a "journalistic rendering", too scarce in details. The second link is the abstract.
      Let's hope the arxiv preprint [arxiv.org] will be good enough.
      • "Good enough" depends on how you define it. It seems that the arxiv version has at least the central theorems as mathematical expressions, and most of the images / figures the Nature paper has. Good enough for me to get a start on this. I am actually working with an R & D organization specializing in this field, so I suppose they'll let me buy a copy from Nature - even if I despise the whole Elsevier / Nature paywalling practice. Now I'll be off reading :-)
    • Re:Paradox? (Score:4, Informative)

      by c0lo (1497653) on Wednesday July 17, 2013 @03:16AM (#44305999)

      (Are those genes selfish? Because if so, this) may be a classic case of resolving a Braess's paradox [wikipedia.org] by removing a trigger for selfish behaviour.

      It has to do with non-linear systems that have many points of equilibrium (Braess's paradox involves another example of the same, except the equilibrium is considered in the Nash sense).

      A quotation from the arxiv paper [arxiv.org] that says what's all about:

      For example, a damaged power grid undergoing a large blackout may still have other stable states in which no blackout would occur, but the perturbed system may not be able to reach those states spontaneously. We suggest that many large-scale failures are determined by the convergence of the network to a “bad” state rather than by the unavailability of “good” states.

    • by Dishevel (1105119)

      (Are those genes selfish?)

      They all are. By Design [wikipedia.org].

  • An analogy (Score:5, Insightful)

    by ardor (673957) on Wednesday July 17, 2013 @02:57AM (#44305961)

    Reminds me of software bugs which are "fixed" by disabling subsystems around them. Example: in a media player, AAC playback sometimes freezes and causes glitches. Solution: disable AAC playback, ensuring that the media player does not reach this undefined and broken state.

    • Re:An analogy (Score:5, Insightful)

      by Anonymous Coward on Wednesday July 17, 2013 @03:20AM (#44306021)

      So, in the networking sense, he as discovered route poisoning. [wikipedia.org] I bet he will soon discover more fascinating things such as the firebreak [wikipedia.org] to fix vegetation being burned or the embargo [wikipedia.org] to fix uncooperative nations by refusing to cooperate with them.

      It's not that I devalue the work. The work is good, its just that math is only a descriptive model yet is often given the accolades of being causal.

    • My analogy is my cell phone. Sometimes I just wish it would just "lose" it's Sprint service and go into roaming mode, so my data would work. :(
      • by jimicus (737525)

        Most of them do that automatically - they'll automatically try to authenticate against the strongest signal they can find that will let them authenticate, no matter what network it is.

        When your provider has a roaming agreement in place with another provider, they set things up so that they can authenticate the phones of the provider with whom they have the agreement. Most providers don't enter into roaming agreements with other providers within the same country, which is why your phone doesn't usually start

        • by jeffmeden (135043)

          Most of them do that automatically - they'll automatically try to authenticate against the strongest signal they can find that will let them authenticate, no matter what network it is.

          When your provider has a roaming agreement in place with another provider, they set things up so that they can authenticate the phones of the provider with whom they have the agreement. Most providers don't enter into roaming agreements with other providers within the same country, which is why your phone doesn't usually start roaming when you are in your home country. But this can mean that if you live near the border of two countries and you pick up a stronger signal from a foreign network, your phone automatically jumps on it.

          The first you learn of this is when your next phone bill arrives as a multi-volume set of books, luxuriously bound in finest Italian leather and delivered by a liveried courier.

          The US is a tad bit... Different. We have 3.794 million square miles (to your 1.6M sq miles of EU) and mobile subscribers, god bless them, expect their shit to work _Everywhere_. Sprint especially does a good job of borrowing access on other networks (mostly Verizon) since Sprint isnt nearly as thorough in terms of coverage.

          A US provider not entering into a roaming agreement with another US provider would be, eh, a little bit ridiculous. Sprint piggybacks Verizon, and Tmobile piggybacks AT&T, each of

    • by zildgulf (1116981)
      Believe it or not I have done the same to the network when something goes wrong. We had real problems with Multicasting in our network when we upgraded our switches. I found out that it was easier to disable multicasting on our switches, which by default treat multicasts as broadcasts. Since we were using multicast only for Norton Ghost, and not for media host streaming, that was a perfectly acceptable solution.

      The lesson: If a feature malfunctions the fastest way to fix the problem is to delete the
  • by Anonymous Coward

    stupid windows 8.1 can't buffer a flash video enough to watch (without freezing)on a slow connection no more.

  • And then the next time something break, just break even more.
    I'm sure that system will be sustainable for decades without ever encountering any issue that can't be solved by breaking yet more.

    • by azalin (67640)
      Well once the stable state is reached, there is no more need for any additional breaking.
  • Broken leg? (Score:5, Insightful)

    by mjwx (966435) on Wednesday July 17, 2013 @03:38AM (#44306069)
    So if I have one broken leg, the best solution is to break another leg and maybe an arm?

    Mechanical and electrical problems are radically different to biological ones as they dont self heal/mutate.

    With a cell, you're attempting to force it into a reaction by breaking it more. We do this because we dont have the knowledge or experience to fix it ourselves. With networks it's the opposite, isolate the damage, route around it if need be and then fix the broken components. Yes that's a simple view, but the basis for fixing network issues.

    If my route to Sydney is down, deliberately breaking my route to Melbourne wont help if there is a physical cable problem or some idiot down in the NOC changed the route cost to 10000 on the router. Nope, instead of one route to Australia down, I now have two.
    • Re: (Score:2, Insightful)

      by Anonymous Coward

      The point here probably being, that if your route to sydney is misbehaving, losing packets, being otherwise slow etc. you break it even more from the point where it's misbehaving and the system should reroute it around that point.

      • by mjwx (966435)

        The point here probably being, that if your route to sydney is misbehaving, losing packets, being otherwise slow etc. you break it even more from the point where it's misbehaving and the system should reroute it around that point.

        Not really, Don't break, isolate. If packets are going faster through the Melbourne link, the router should pick that up (lowest route cost), if worse comes to worse I can raise the cost of that route. Breaking it more doesn't help, if there is a router misconfiguration, smashing the fibre with a sledgehammer will do nothing to help me. Instead of just a misconfigured router, I now have a misconfigured router and some broken glass to fix.

        The closest thing we have here is replacing suspected bad with kn

        • by MrP- (45616)
          "Breaking it more doesn't help, if there is a router misconfiguration, smashing the fibre with a sledgehammer will do nothing to help me."

          But it feels good!
      • by kasperd (592156)

        The point here probably being, that if your route to sydney is misbehaving, losing packets, being otherwise slow etc. you break it even more from the point where it's misbehaving and the system should reroute it around that point.

        Exactly right. And this should come as no surprise to anybody with experience in distributed systems. Often a partially functional component cause more damage to the system than a completely broken component. This is because it is much easier to design a system to deal with failu

    • With a cell, you're attempting to force it into a reaction by breaking it more. We do this because we dont have the knowledge or experience to fix it ourselves. With networks it's the opposite, isolate the damage, route around it if need be and then fix the broken components. Yes that's a simple view, but the basis for fixing network issues.

      Assuming they're talking about fixing the physical aspect.

      Now, if you dealt with a broken politician by removing more of them from the political system it just might work.

    • by azcoyote (1101073)
      Exactly--this claim is overly generalized to fit all 'networks.' The problem, then, is that this is less science and more philosophy. If its claims are valid, then its applicability to any particular network needs to be heavily filtered through practical considerations and common sense. But then, we still have to justify in the first place the use of a broad (and even metaphysical) model as determinative (and not merely illustrative) within empirical science. I'm not sure such a justification can be made on
    • by zildgulf (1116981)
      What they really mean is that if a feature of a module is malfunctioning or broken then take out of production and/or disable the entire module.

      Even though your leg's skin, joints, and muscles are fully functional the best way to handle a broken bone in your leg is to pull that leg from production. In other words, if only your tibia is broken, the best thing is to stop using the entire leg until it is healed instead of trying to find a way to continue to use the leg without putting weight on the tibia be
    • by romons (2767081)

      If my route to Sydney is down, deliberately breaking my route to Melbourne wont help if there is a physical cable problem or some idiot down in the NOC changed the route cost to 10000 on the router. Nope, instead of one route to Australia down, I now have two.

      Hmmm. If your route to Sydney went down momentarily, and your router decided that the best way to get there was through Tasmania, and it takes 20 minutes for the routing system to figure out that Sydney came back up, then yes, it might be that after Sydney comes up again, killing your route to Tasmania would make things better overall, because it would force the system to figure out a better way.

      Another example would be an optimization algorithm that picked a local minimum. Breaking a link might enable the

  • Finally (Score:5, Funny)

    by Chrisq (894406) on Wednesday July 17, 2013 @04:02AM (#44306155)
    An explanation of why they followed XP with Vista

  • This guy basically came up with "damage mitigation" and a way to highlight a "critical path"? -essentially if something is broken to disable everything that relies on the something and carry on in limited capacity.

    The model might be perfectly workable but the ambiguous "undesirable state A" etc is going to be nearly impossible to implement.
  • by wisebabo (638845) on Wednesday July 17, 2013 @04:16AM (#44306201) Journal

    So if one considers the weight distribution the bulkheads in the ill-fated Titanic as a "network" perhaps it would have been possible to save her (or at least keep her afloat long enough for the Carpathia to rescue all the passengers) by further "damaging" it.

    While it has been often said that she could withstand any two bulkheads being flooded, in truth she could take many more, various simulations show that she could take at least four being flooded, in various combinations. And this was with them being COMPLETELY flooded (up to the top of the bulkhead partitions); if she were on a more even keel they would only flood to the water level.

    The problem of course is that the Titanic was NOT on an even keel. When the compartments, all in the front, were flooded that caused her to pitch down. The water kept rising until it went OVER the bulkhead partition, flooding the next. This caused the weight in the front to increase even more which caused her to pitch even further and ... you get the picture.

    So, thinking of this like a damaged "network"; perhaps if the captain had flooded one of the far aft bulkheads (breaking the network more), the Titanic would not have pitched downward as much and the water wouldn't have overflowed the bulkheads (they were not watertight, water could go over the partitions). This might have prevented the cascading effect which led to the sinking of the world's largest ship just two and a half hours later.

    Would she still have sunk if they flooded one of the aft compartments? Maybe but it might have happened much more slowly and gently (no scenes of people falling down a nearly vertical ship!). And if the downward pitch was reduced so much that the water didn't surmount the bulkheads (the partitions separating them were quite high, much higher than the normal water level), maybe she would have remained afloat!

    The builder (designer?) of the boat was on her when she sank, I wonder if he considered this? Or did the thought of damaging the boat further never cross his mind?

    • Sorry I commented on this thread as I'd mod your excellent post UP -

    • Perhaps moving them all aft would have been sufficient to keep the boat balanced too !?

    • by ledow (319597)

      It seems a pretty obvious and common sense statement, I don't get why it's worthy of an article.

      When your network falls into a broadcast loop, how do you "fix it"? You break the network. When your website is overwhelmed and the traffic is killing your other, necessary external connections, how do you fix it? Turn your website off and let people bounce off a 404. When your car oxygen sensor is broken and won't let you run the engine properly, how do you fix it (at least in some models) - disconnect the s

    • The builder (designer?) of the boat was on her when she sank, I wonder if he considered this? Or did the thought of damaging the boat further never cross his mind?

      Neither. He was all qq because his beautiful fiance was banging some street yokel from steerage.

    • Re: (Score:3, Informative)

      by slimdave (710334)

      Yes, deliberate flooding was recognised at the time as a solution for accidental flooding problems.

      Taking the famous Andrea Doria collision with the Stockholm, there was an attempt to flood empty tanks on the former in order to right the list caused by flooding that followed from severe collision damage. To no avail, however.

      http://en.wikipedia.org/wiki/SS_Andrea_Doria#Assessing_damage_and_imminent_danger [wikipedia.org]

    • Sorry I don't understand. Do you have a car analogy?

    • by zildgulf (1116981)
      Actually that makes sense. Purposeful strategic flooding, further breaking the Titantic, if it were possible, could have saved the Titantic long enough to evacuate the passengers on another ship because the weight of the water in the ship would have been distributed more evenly. The Titantic would have sunk anyway but probably with far fewer causalities.

      Actually in World War 2 the Navy did this all of the time. If the warship was in danger of capsizing in a storm the best thing they could do is to actu
  • So if it's true, can we apply it to every other aspect of our lives? How about applying it to a society? If there is something bad about the society, the only way to fix it is to make it worse, by taking out more goodness from the society, until it reaches a point where it's 'good' again? Who defines what's good and what's not in this case?
  • Cornelius's model implies breaking the system down until there is only five elements left.

  • If this guy had ever worked IT, he'd know that if you have a serious problem that needs to be solved NOW, and you are already dead-in-the-water, this is 100% standard practice, and something that is self-evident without the help of a mathematical model.

    I used to work Enterprise-level tech-support for a large maker of computer equipment, and our first response to a customer that was complaining he was essentially out of business until his computers were back online was to request he reboot plausible sources

    • "Hold on $insert C level executive here, I'm going to break more shit, trust me, it's for our own good" You're fired.
  • I don't see the practical application of something like this in a physical networked environment. Thinks like this may sound nice for permastudents' theses, but may or may not have a practical use without one hell of a architectural and best practices shift rather than just replacing a failed piece of equipment or rectifying a software issue and closing the ticket?
    • but it detects carrier just fine, so if the hardware is only checking for carrier and not actively testing throughput it won't detect a problem.

      The solution is to shut down the link totally in order to force the failover mechanisms to pick a different link.

      • You would pick that up in your normal troubleshooting routine though, swapping the card/cabling, but I see your point.
  • This is the same basic logic that's employed by Skynet and most other fictional rampant AI.
  • by gmuslera (3436)
    It could work. Internet is broken as US government decided that noone (at least, non-americans) deserves privacy. Now consider privacy (at least, whatever you in private) as intellectual property, then decide that noone (at least, american) deserves intellectual property, and you will find plenty of corporations forcing US to consider going back in the no privacy idea.
  • we do this at the board level on h/w all the time when manufacturing h/w (PC boards, graphics carss etc.). If a failure is not obvious or intermittent you change things until it is fully broken and easier to find and fix. It is especially useful in engineering investigations.
  • Makes perfect sense. If it is too broken to fix - refactor it.
  • Is this how chiropractors work? I just spent two hours in electro-therapy and spinal re-adjustment. Break the network to fix it, indeed!

  • by sjames (1099) on Wednesday July 17, 2013 @01:28PM (#44309975) Homepage

    TFS and to a degree, TFA missed it. They don't claim that they have suddenly discovered this property, they claim new analysis tools that help to determine what to break and where to break it to return a non-linear network to some reasonable stable state.

    For example, if the power grid suffers a de-stabilizing loss, what needs to be done to move it to a stable state rather than allowing it to go into a cascade failure. We have some tools for that already, and in fact, power is re-routed and load is dropped all the time in order to stabilize things. Better tools suggest either getting to the answer faster, finding the least damaging way to reach stability (the least customers dropped) or better certainty that the actions taken will prevent dropping the whole grid.

    Nobody is claiming it's OK to leave it in that state. In fact, the same analysis tools might help to make sure that the repairs don't inadvertantly destabilize the grid again as parts go back on-line.

  • I think I heard somewhere (early 1990's IIRC) that someone did a study with a neural network where they randomly disconnected nodes (I can't remember what type of neural network it was). At first, the output was gibberish, but when more nodes were disconnected, the output resembled things that were some of the first things the network had learned. Someone likened this to Stanley Kubrick's 2001: A Space Odyssey where when HAL was gradually being shut down, HAL recited "Mary Had a Little Lamb"

"Don't discount flying pigs before you have good air defense." -- jvh@clinet.FI

Working...