Slashdot Mirror
What Happens When 99% of the Net Crashes?
Sara Chan writes "The Internet remains connected on a global scale even if a randomly chosen 99% of its connection points break down. It is, however, in danger if its most highly connected points are selectively knocked out. Recent computer simulations have shown that the Internet is fairly resilient because it is scale free. The latest work, published in Physical Review Letters strengthens this conclusion. Two independent groups of researchers applied percolation theory. Percolation theory deals with systems containing points ("sites") and connections between them, and it analyzes the behavior of the system when some of the sites or connections are removed (it was developed by geophysicists for estimating how much oil could be extracted from reservoirs in a porous medium). Abstracts of the papers are available here and here."
First, our work deals with resilience of many kinds of networks, not just the Internet, and may be a better model of some of the others than it is of the Internet. As several people have pointed out, preserving connectivity in the Internet is kind of pointless if the performance is appalling, as it would be if you destroyed most of it. (We did actually make this point in an earlier version of the PRL paper, but it got cut out to make way for other things.)
The whole paper is available here, and if you look at it you'll see that we also talk about applications to things like the power grid, and contact networks which result in disease transmission. There are also many other important networks which one could apply it to. Distribution networks, like UPS or regular mail; transportation networks, like airline routes or highways; food webs; neural nets. One of the most interesting, perhaps, is not the Internet, but the Web, which is also a network (of pages linked together by links). This has also been shown to have a scale-free degree distribution (A. Broder et al., Computer Networks 33, 309 (2000)) and so should be highly robust in the sense that you can still surf from one place to another even if most of the pages in the Web disappeared overnight.
To answer the question about non-random failures, yes, things are very different there, and this is also discussed in the article. If you attack only the most highly connected domains/web sites/people/airline hubs, etc. you can destroy the operation of the network in no time at all. These networks are very vulnerable to directed attack. This could be very bad, but it some cases it is good. For instance, it means that a focused attack on a disease-causing contact network, for example through vaccination of the people with the highest number of contacts (the so-called "core group"), could prevent an epidemic with comparitively little effort. (Of course, identifying the core group may not be trivial.)
The guy who made the point about the Australian fiber dying last week is onto the right idea. This is the same effect - if you take out the right connection (or wrong, depending on your point of view), then you can do a lot of damage. It's only if you take out a random one that the network is robust.
All the best,
Mark.
As I understand it, the Internet wasn't "designed" to deal with nuke aftermath. It's a wives' tale with a grain of truth deep inside to make it sound good.
The communications system of DARPA may have maximized the advantages of routing, so that .mil sites would have enough redundancy to still talk if a node or a few went down. Computers were just as unreliable then as they are now. Some stay up for months, others bluescreen or blow tubes weekly.
However, DARPA is the Internet . As the name implies, the Internet is an amalgam of many smaller networks. Thus, the whole Internet may not share the fault tolerance that DARPA enjoyed. It's not even clear that the .mil network even still has such traits.
Big business is building most of the Internet infrastructure now. While it costs lots and lots of money if a cable gets cut, corporations still don't have the long-term vision attitude it would take to put in enough redundancy. Dig two transcontinental trench projects with two cables, when you could just dig one? That's a cost.
There's relatively few undersea cables. If there's maybe fifty crossing the North American continent, there's only ten crossing the Pacific. Satellites offer long slow alternatives, but if a deep cable is cut, it's BAD.
[
I remember working for USWest, and large sections of MPLS (as we called it) going out for no known reason quite frequently, although our guess at the time was because whoever was in charge of 204.147.80.1 had taken off the internet software so they could play Frogger.
Hopefully I didn't put any [] around my words.
Something similar happens to alt.* newsgroups when they are not propagated properly, due to the groups not being discussed at all. Some sites create the group, most of the rest don't, then the sites which have created the group are not connected at all, due to the sites in the middle who refuse to forward postings from the group.
The result: a similar situation to what is mentioned here. Perhaps worse, as many don't have a clue of what has happened. An example of this is alt.tv.sliders.creative. On some ISP's you can read some messages, but on other ISP's on the other side of the world, you can only read OTHER messages, and never the two shall mix!
SCIREV.NET - fanfics,reviews & more
Take out the root.a (etc...) NS. then watch 99% of the population not know what to do!
Main sites would go down in min because their ttl's are set so low.. someone does a look up.. oops not cached anymore.. nowhere to get the authorative ns from.. no ip resolution..
this would stop the web dead in a matter of hours.. and since 99% of people don't use anything other than their AOL browser. gameover.
If 99% of the backbone to the 'net is down (due to nuclear war, supervirus, whatever) don't you think it's reasonable to assume that 99% of the PC's connected to the 'net would be down as well. I highly doubt that Joe AOL's computer would still be around after a disaster major enough to crash 99% of the Internet's bandwidth. Seems to me it's a moot point to begin with.
-atrowe: Card-carrying Mensa member. I have no toleranse for stupidity.
Then, by my calculations, we'd all party like it's 1989...
I've got a remington 870, a sig P220, a colt AR-15, some body armor and a whole lot of ammo. If I can't get through to my favorite site somebody will pay.
I am not fucking joking about this.
--Shoeboy
Slashdot would post stories like "Ping returned in West-Coast dial-up pool, human involvement suspected." and then everyone on Slashdot would try to ping that IP, and the router would go down, and the story would be updated with "False Alarm, probably just an automated hourly windows update."
Lather, rinse, repeat...
WARNING: there is a trojan on your
Fault tolerance, yay!
Comment removed based on user account deletion
theres 99% less people who can compete against me getting the first post!
,
faeryman
How to destroy the Internet:
- Acquire dialup account. Stolen if possible so that they have less contact information for you on file. For best results, see if you can find an ISP that does not record remote calling IDs.
- Acquire a shell account or preferably run Linux (free) on your own computer.
- Read bugtraq. Learn the skills of the skr1pt k1dd1e. IRC can help too.
- Start 0wning servers. Preferably servers with foreign domains. The language barrier helps and less mainstream countries have a much harder time dealing with internet problems. Also, it helps to own one server and then own other servers from that owned server. If a sharp admin tracks you down, you simply need rm -rf / the owner server to cut off all associations with you.
- Start sweeping for broadcast addresses. netscan.org also publishes a handy list.
- Take advantage of all of the distributed denial of service tools out there. Start outfitting your shell accounts with said tools.
- Set up for the big day. Also have a backup plan ready and tell no one about what you're going to do. Your friends will easily betray you for the glory of bringing the internet killer to justice. The thrill may be overwhelming but a lot of people will be mad at you if they find out. If you can't keep a secret, don't do this.
- Pick the big day. New Years Eve would be conveniant because people will be drunk and confused and think of the Millenium Bug.
- When the big day arrives, your goal is to hit the root servers. Hit them with everything you possibly can. X.root-servers.net. Preferably send a barrage of spoofed, fake DNS requests. Spoofed so that they have no hope of hunting down the machines you're doing it from (or at least all of them, in any timely fashion) and fake DNS requests because they cannot function if they filter DNS.
- Never connect to these machines again. Destroy your own machine with fire.
The internet should now be unusable to about 99.99% of it's users.My research is just as valid as theirs, and mine is much more plausible. So pfft.
I have access to the full article and gave it a read. Its a perfect example of why physicists shouldn't write articles about the real world. They are assumed that all nodes of the network are equal. This is most definitely not the case with the Internet. There nodes upon which others depend. And, as if that wasn't enough, there are also choke points in the real Internet. I liked the LA Freeway analogy that another poster gave. Let me elaborate on this analogy. Half the traffic, maybe more, takes the Harbor Freeway to the Hollywood Freeway to get out to the SF Valley. The other half or so use the 405 to the West. The geography breaks down so that if you live SE, you take the Harbor, Hollywood. If you live SW you take 405. Now, the interchange from Harbor to Hollywood is a narrow two lane affair left over from the fifties. If that goes down from an accident, no one on the east can get to the valley. The same thing can, but is less likely, to happen on the Internet. Take out Downer's Grove and one or two other spots, and whammo, the Internet is segmented and cut off.
In crisis cases, it would help immensely if packet prioritization/QoS were applied. If a pipe is so full that it is losing packets, the ones to kill off are the expendable ones -- web traffic, random high-port stuff, etc. If the gigaswitches are able to make apropriate sacrifices under duress, leaving routing messages, ICMP, etc, to survive where port-80 does not, the ability to be self-repairing is likely much improved.
Actually, in L.A., where it really IS UnAmerican to do anything except drive, the busiest freeway in the world (I-10, the Santa Monica Freeway) was shut down after the Northridge quake. We routed around it. Traffic was even slower than usual, but it got through.
General Relativity: Space-time tells matter where to go; Matter tells space-time what shape to be.
...after all, that *is* what it's designed for!
That way, if nuclear war breaks out in the US, you can still send a message to the other side of the country that says "NE1 HERE??? A/S/L???" on IRC...
---
pb Reply or e-mail; don't vaguely moderate.
pb Reply or e-mail; don't vaguely moderate.
... my latency on Counter-Strike is awesome!
I'll teach them razzafrazzit LPBs...
Jay (=
This is a theoretical breakdown of the internet assuming that every node on the network has a physical link to more than one node. In practical terms this is bullshit. The internet is a collection of everyone on the network (including us folk dialing into a server hooked into a a larger portion of the network). If someone EMP's the local dialup and you don't have any backups, you've lost a bunch of nodes. If you EMP a MAE you've just disconnected another shitload of nodes off the network proper and for the most part split that severed part of the network into a bunch of small networks. Over the distribution of the internet there comes a series of physical locations a good portion of your data is routed through. Turn even a handful of these puppied off and you fragment the internet into a bunch of littlenets. So technically this still counts as "internet" theoretically but in practical terms you're shit out of luck. Speaking practically, all you need to do is cut a bunch of trunks or fibre lines and suddenly a bunch of networks die, not to disconnection but the fact that for many reasons they set their packet TTL really low. When these packets end up routing through a couple of bottlenets they will end up dying. If you want to demonstrate this go to your company's network room. Start yanking plugs out of the sockets. If you can yank the ones out that belong to the file and print servers or just dump a can of coke on the router/switch/hub itself. About 30 seconds before you're fired after an intense bout of screaming by your boss, explain to him you were testing out a theory. You just created an experiment sample, and disconnected 99% of your network. That one plug you didn't unplug is still going strong though! Your pr0n download will finish and you can pack you Zip disk with the rest of your stuff. At least the theory applied well to the real world.
I'm a loner Dottie, a Rebel.
the load on the few remaining servers if most of the /. community was included in the 1% that remained standing?
I've thought about this a few times and wondered what effect the portalization of various markets would have on the stability of the 'net as a whole.
Some companies pay $$ to have traffic routed across their networks, what happens if their network gets toasted by some super-malicious script kiddy or some super-malicious asteroid? How redundant is the 'net now-a-days? A lot of traffic routers through LA and San-Fran (I ran a tracert on my home connection from work, [I live in Washington State] and it seems my packets route through california... You'd think there'd be a shorter way to travel 20 miles...). What happens when Cali slides into the ocean after a 9.99 quake? Then again I think there'd be more important concerns if that happened, but still, in the aftermath I'm sure some of us would be jonesin' for our 'net connection.
When you get to point of having 99% of the internet down, don't you think that whatever caused it (bombs etc.) will have messed with the power supply?
Don't be shocked, fellow reader, but i'm afraid that your average joe doesn't even know what a UPS is, so i'm guessing there may be a shortage of web users at that point. Not too many people are going to experience a nuclear war, and then ask "I wonder if slashdot is still up?".
Drag n' Drop DVD Recommendations
I remember someone did a ghetto fire under some trunk connection near a bridge, and Minnesota was out of Internet connectivity for a whole day. Whoo hoo!
An anogolgy: You could probably destroy lots of surface streets in a large city without affecting the entire city much, but if you knock out a freeway you're fucked. Same with the internet.
Besides, if 99% of the nodes go down, surfing out to read slashdot probably isn't going to be at the top of my priority list... I'll most likely have my hands full fighting off alien invaders or trying to find shelter from the massive meteor storms that are destroying the world...
The full article is available without subscription to PRL ... you can get it from the lanl archive here.
How about the '88 Internet worm?
General Relativity: Space-time tells matter where to go; Matter tells space-time what shape to be.
It is.
Mode (3) smart-aleck mode. Press * to return to main menu.
You mean like the root nameservers? What are there, 12 nowadays? Hmm...
We all seem to forget that the whole commercial internet would come to a screeching halt if those twelve servers--including any backups that might be in place--all decided to crash at about the same time. Minimal chances, I suppose, but the number of root nameservers is a very tiny fraction of the number of hosts on the Internet.
Granted, the connectivity wouldn't be there, but I'm have no idea what the IP address for the Slashdot web server is, and I'm not going to look it up. If the nameservers went down tomorrow, I'd have connectivity to my school and my ISP, but that's about it. Most of the web would be absolutely useless to me.
Do we solve this by distributing DNS? I have no idea. I can't believe that distributed DNS works. This danger of putting all our Internet eggs into twelve root nameserver baskets may be a danger we just have to risk.
The only policy I would implement if I were ICANN has nothing to do with risk, but opening up TLDs to everybody. Why can't they sell TLDs like they sell subdomains now? I'd love to own the .commandant TLD. Plus you could be a lot more creative and natural in the naming of websites. For instance, you could go to staroffice.sun instead of sun.com/staroffice, or netscape.navigator to check the latest netscape browser.
That's my $0.02 about TLDs and root nameservers crashing.
I do not belong in the spam.redirect.de domain.
Googlehas 1,326,920,000 pages indexed. I'm sure there's more than that.
-atrowe: Card-carrying Mensa member. I have no toleranse for stupidity.
yes, you guessed it... 99% of all text == "99% this or that" :)
Maybe an idea is to have a `slow-but-sure' setting that routers don't
have to deliver quickly, but try to ensure that they aren't dropped.
This would be a bad setting for urgent content delivery, but it might
be a good setting for spreading information about traffic spikes, etc.
Maybe "on a global scale, the internet will remain connected", however, as a user, my ISP has one or a few connections "to THE INTERNET". If that node breaks down (here in the netherlands: "the amsterdam Internet Exchange") I'm cut off from the net.
So, what use is a "globally connected net" if 99% of the users are cut off, as well as 99% of the web servers?
There is a 99% chance that google is cut off, so that you can't search the net with google anymore. If you're lucky, google still works, but returns 99% broken links.
Maybe from a theoretical viewpoint there will still be a link from Europe to America, however, "the net" will be completely useless.
(argument: but many ISPs have more links to the internet than one. Sure, ok, but there are more nodes than one involved before you get a useful link through a secondary link. In the end I estimate that still chances are less than 1% that you'll be connected, even if you have an ISP that has multiple redundant links....)
Roger.
"After a nuclear war, there's only three things that will surive: Cockroaches, Lawyers, and Spammers."
Standard I/O Error. Incompetent/Operator.
They proved they could do it in less than 30 minutes.
More race stuff in one place,
than any one place on the net.
>ocean after a 9.99 quake?
A few months ago there was an article, an "Ask Slashdot" I beleive, that asked the question: "what would happen to the net if the US dissappeared?".
The results were somewhat supprising (once you got past a lot of the crap)....
Now, by and large, we got a lot of the standard, "good riddance to the lazy fucking arrogant "USians"" trolls. But SOME people had enough clairity of mind, and desire to contribute to actually post trace routes. And I found the results quite astounding.
Austrailian packets tracerouted through San Jose to get to New Zealand. Packets from London were going througn Boston to get to France (!!!). Vancouver packets went through San Francisco and Boston to get to Toronto. Hell, one Australian, IIRC, had a traceroute go through San Jose just to get his packets from one Sydney ISP to another!!!
So to answer your question, if California crumbles into the Pacific, I think that much of the rest of the world goes into the shitter with us.
Now, much of the REST of the US, I doubt would be missed, or, indeed, it's absence noticed. (I, for one, am still hoping for a bigass comet to take out texas... (before January 20, please)). But the Bay Area and New England ARE rathar critical.
So, as long as San Francisco, Boston, and the appendage cities that surround each, survive whatever apocolypse that wipes out 99% of the world, I think we'll be just fine. Take out either or both, and the results will NOT be pretty.
john
Resistance is NOT futile!!!
Haiku:
I am not a drone.
Remove the collective if
Imagine all the people...
With my ISP? I might not even notice for a few days. Dang, I gotta get DSL.
---
Good judgment comes from experience.
Experience comes from bad judgment.
Isn't sudden reversal a matter of a couple decades where we go without magnetic poles?
If we have no magnetic poles, not only does our electronics fail but so do humans due to radiation.
Keep in mind its the magnetic attraction at the poles that drag the various radiative particles from the sun away from the human populus. Can you say deep fry in 2 minutes when outside?
Rod Taylor
You know if you want avoid earthquakes move to Minnesota. The last one we had rated a 1, occured 20 years ago and was the reactivation of a 1.5 billion year old fault. Granted Minnesota has evidence of a fault system that would dwarf the stuff on the West Coast, but the last time it was doing anything was 1.5 billion years ago.
Impersonating Tycho from Penny Arcade since before there was a PA.
It would work, but not with the current pricing schemes.
The point is, to avoid the scenario you're suggesting, there needs to be a deterrent against marking all traffic as high-priority. It could be done as placing network limitations, as it is done with IPv4-TCP out-of-band data (out-of-band data is a mechanism to send "urgent" packets overriding TCP's congestion-control mechanisms; said limit consists in only allowing one packet of OOB data to be alive in the network at a given moment), but it would probably be quite expensive to enforce. Or it can be done (much more effectively) using billing: "you can mark your packets high-priority. We charge by the byte, and high-priority packets cost twice than 'bulk' packets".
This is not, as you can see, a technical issue, but a marketing one. Unfortunately the current dominant pricing scheme is flat, which offers no such deterrent, indeed it's quite the opposite.
Try xxx.lanl.gov for the preprints.
Mode (3) smart-aleck mode. Press * to return to main menu.
Remember that these routes are just the most preferable routes at any one given time. It doesn't mean that they are the *only* routes. If San Jose died, ymight find Australian packets going via Singapore, Hong Kong, South Africa, then Norway to get to New Zealand. Just because the best route is geographically a long way, it doesn't mean it's the only route.
perl -e 'fork||print for split//,"hahahaha"'
" ... the powerful percolation-based approach may help Internet architects to maximize resistance against Internet attacks, by controlling the distribution of nodes having certain numbers of connections."
Rip storms. It happened to MCI back in 95 and brought down their whole network.
More race stuff in one place,
than any one place on the net.
You and that hot chick from high school, she still wont go out with you.
How many nukes would a missile defence umbrella stop and at what efficiency?
0, the defense department has no nuclear missle defense umrella. Give it 500 billion dollars and it will be happy to build you one with 100% efficiency right away.
If I was cleanly severed in half and survived, how long before my body dies from being unable to relieve itself?
About a week
What quantity of LSD (insert any other drug) would be needed to get an elephant high?
It is unknown wether elephants experience highs the same way humans do, but if they do, a dose of approximately 10 times the normal human dose would do it
If Microsoft finally developed the perfect OS, what would it's next killer app be?
People ask me if I'm afraid of earthquakes here in CA, I'm not too afraid but chances are they should be.
Tony
The problem with prioritization is that it works great in theory, then you put the scheme in user-space and it falls apart. What's to stop users from applying a high-priority to their web traffic?
:)
If the users control it, the best you can hope for is "Geeks who know how to set IP Precedence fields get to go first!"
On second thought, maybe that's not a bad thing, after all.
Seriously though, unless you're policing the priorities, it's impossible to make this work in the real world; and if you *ARE* policing the priorities, you're more likely to introduce latency that will degrade performance when there isn't a crisis.
*sigh*
_________________________________
Now!
What happens is 99% of the net crashes?
It makes me really happy that I am part of the 1% that does Not use AOL.
The anti-salmon
If two men in a boat can sail up to the side of a miltary vessel and blow a fucking great hole in it, I'm sure that if anyone that dedicated wanted to, they could make a real mess of the Internet. But physical damage is going to be reasonably easy to fix as most major net nodes should have disaster recovery plans in place.
To really make a mess, why not write a nice virus or worm that would be much harder to react to and recover from.
Of course the ultimate would be to combine a few pieces into one large puzzle :- mass client infections, Root DNS DDOS attacks, email hijinks, and take out a few key cables/bottlenecks with backhoes. The trick is to create cascading failures that individually could be fixed, but the presentation of all problems at the same time makes the response and recovery that much more difficult.
Best Practice dictacts that anti-virus and firewall vendors get hit as well, just to highlight the point.
PhysRev is Slashdotted!!!! Yay!!!!!!!
Mode (3) smart-aleck mode. Press * to return to main menu.
Do you have any reliable links to back that statement of yours up with.
It seems a little strange to me that a communications system that was designed to allow remote users to communicate in the aftermath of a nuclear war would be so fallable.
-atrowe: Card-carrying Mensa member. I have no toleranse for stupidity.
Granted the net has changed since I last did Netop work, but there's still some practical concerns that I dunno if the article addressed. Yeah, globally the net is up with 1% of the routers going, but locally you're screwed. Here's a couple of cases in point. Several years ago there was a fire at Downer's Grove in Illinois. Downer's Grove besides being a suburb of Chicago is home to a large portion of the telecommunications interconnections in the country. From my view, it seemed as if Downer's Grove connected every thing east of the Missippi to the western part of the country. Sorta like the corpus callosum--I know I mispelled it-- in the brain. So the fire took out a couple of MCI's backbone routers--more than that. From my point of view it looked a whole lot like the country had been split. The west was fine. The east was fine, but when the two tried to talk there was nothin going through. It was a mess. No one knew why, what or where. I dunno if it is still possible to have this kind of break, but I have to wonder if the article takes into account the phsical topography of the net vs the network topography. The fact is that if the net still runs the way it did, and I'm pretty sure it does, the distributed nature of the net is misleading. Truth is that the major infrastructure is piecemeal centralized in physical locations. I'm talking about places like Downer's Grove, Arlington, and Denver--I think. These cities support a lot of the net's critical backbone routers. What's more an awful lot of them are centralized in single POPs. That means they're in a single building a lot of the time. So, when we say there's connectivity with 1% of the routers, I have to ask what if we still have connectivity with 99% of the routers. It all boils down to which ones. This incident was Octoberish '94 for those who wanna go digging through trouble ticket archives. Another incident happened in like 92. There was a bug in Gated that caused a routing problem in Boulder to propagate all across the NSFNet backbone. Everything went down. A few of the more robust machines came back up on their own and the net lurched forward, but for practical intents, unless you had a big 'ol cisco connecting your lan, you were off the net. Anyway, I should actually read the article before shooting my mouth off. Besides I'm sure a lot of people can find other cautionary tales by digging through their trouble tickets.
What was I thinking? Add two to the years cited. Damn not enough caffeine.
If 99% of the net went down, we'd all learn to appreciate the 1% of pr0n that's left.
Forget Ted Kaczynski, you can become a much more effective terrorist.
1. Build a EMP "bomb" that consists of several farads worth of capacitors and a big solenoid. Have it activated by a timer, remote control, or (if you're particlarly fancy, put an embedded controller like a uLinux simm in there and wire it to ethernet).
2. Mount the bomb in a rack-mount server case.
3. Mount the rack-mount server case in a full-size rack full of "real" networking equipment; servers, dialup servers, and a router.
4. Colocate this rack at MAE-East.
5. Repeat 1-4 for MAE-West, Atlanta, Chicago, etc.
6. Have them all generate massive EMPs at the same time, preferably during a usage spike (Monday at 9AM EST, weekday evenings around 10PM EST, etc).
Extra Credit:
If you're charging the EMP bomb using 220V (which isn't entirely uncommon in colo sites), you might even be able to get a couple blasts off before it melts itself.
-Chris
...More Powerful than Otto Preminger...
Go outside maybe?
Yeah, the lawyers and spammers would probably compete them out of existence.
It's 10 PM. Do you know if you're un-American?
If 99% of the net was taken down, you would most likely be talking a major, global catastrophe, like world-wide blackout by Magneto or something ;) I find it unlikely that even the most organized of terrorists could accomplish such a feat, but I suppose anythings possible. Regardless, it's a major catastrophe.
So if 99% of the web is taken out, what's left to view? If fate has a sense of humor, a bunch of Geocity pages I suppose. But really, most sites are going to be down, so it's improbable that you could get any work or business done over the Internet under such conditions.
I couldn't get to the provided links (Slashdotted already?), so maybe I'm missing the point of the study. My question is, it's nice to know the Net will still function with 99% of it gone, but if it ever came to that, what would be the point of it being up at all?
[shudder]
"It is a greater offense to steal men's labor, than their clothes"
I dread to think of the assumptions that have been made in electrical and electronic equipment that would be affected by this!
As the story presentation states, the net is still very vulnerable if a few highly-connected nodes are selectively taken out of commission. Keep in mind that random attacks mostly get leaf nodes, which have no effect at all on network connectivity, and the rest usually have enough redundancy that there is at least one good path from point a to point b. It may be congested, but then again, if 99% of the net is down, there's a good chance that most people are hiding in their bomb shelters, so you should have it mostly to yourself, just as the original DoD/DARPA designs assumed.
WARNING: there is a trojan on your