Bandwidth Demand at American Universities 392
Robert Rwebangira writes: "There is an article in The New York Times (free reg required), discussing college students 'insatiable demand for bandwidth.' Of particular interest is the continuing prominence of file-sharing (inspite of the demise of Napster) and the amount of bandwidth consumed in even 'legitimate' activities. It seems students demand for bandwidth just keeps growing."
To find the reason.. (Score:1, Funny)
It's a sad day (Score:5, Funny)
In my college days ... (Score:1, Funny)
Jeez. Kids have it so easy these days.
640K Should Be Enough For Everybody (Score:5, Funny)
Of course by then I'll be demanding real-time, life-size holographic video of a "phone-call" to a friend in Asia @ 3 million DPI.
Then finally, matter transport. I wonder how many bytes it'd take to decribe each atom and all its subatomic particles. How many atoms to a human body? Let's do it Star Trek style, and do it in about 5 seconds.
Fast forward a million years, and let's say we haven't blown each other up yet. We'd probably be at the equivalent of God by then.
"Hey Jeff, fancy creating a solar system today?"
"Why not Bob?"
"Well fancy that. OK." *click* "What do you think of that for a Sun?"
"Pretty impressive. Hey let's transport Dave's planet from quadrant four over here. That bastard is always gloating. It'll take him a few seconds to find it."
"OK." *click* "Hey it sort of looks nice doesn't it?"
Re:It's a sad day (Score:2, Funny)
Kierthos
Re:In my college days ... (Score:4, Funny)
But I digress. You young whippersnappers think you have it so good with silicon, you ought to try pushing carts of beads uphill by the bucket uphill both ways with no round wheels.
The Original USC (Score:3, Funny)
Re:Why do gamers have to get scrood? (Score:5, Funny)
I suppose you don't remember the day this fall when packet shaping was turned off [harvard.edu] and absolutely nothing worked. The "poor performance" mentioned in that announcement is quite an understatement -- traceroute probes came back in times on the order of seconds or not at all. I also remember the days just before the traffic shaping was put in place and I was getting over 5-second ping times. People like games, but when email and the web stop working, people quickly start thinking about realistic priorities.
You also might ask yourself: did your games work when no traffic shaping was in place and ping times were measured in seconds and packet loss was rampant? I doubt it. How, then, can you blame the shaping for your problems if your problems didn't go away when the shaping did?
First of all, because of retransmits, dropping packets can lead to high "ping" times, depending on the protocol/application and what it considers a "ping." Second, the software may be trying to "smooth" out the traffic to fit under some limit -- queueing packets from burst periods to be transmitted in lower-traffic periods.
That's a nice idea in theory, but the problem in practice is this: tracking all those individual TCP flows would require immense amounts of computation by the router. AFAIK, we're talking orders of maginitude greater than what is currently available. From what I hear, the really expensive Cisco router at the border is already extremely busy doing just the simple packet shaping, which just limits the aggregate high-port traffic. Breaking that one giant flow into millions of little flows is non-trivial and probably impossible.
Now I'm no networking expert, but it seems to me that doing traffic limitation on a per-user or per-flow basis would probably require some sort of distributed model that did the limiting closer to the user. This might mean not only replacing all the switches with more expensive models, but also hiring new staff (non-trivial) to install and integrate the new hardware into the existing network and to maintain the configuration on all those switches or to write some fancy new automated system to do the maintenance. Of course, all that is just another idea, and there may be some other pratical considerations that make it even less feasible. It also sounds like a lot of work, and considering why you would be asking them to do it, I can imagine that it would end up a lower-priority item than other things that the FAS network people have to do.
I don't know either, but the file-sharing folks have shown themselves to be pretty adaptable. If they would play nice and limit themselves to certain ports and protocols, then everything would be easy. And, of course, who knows whether this would just require too much CPU time, as well.
Anyway, some background/historical info:
The undergraduate dorms get their connections through the FAS (Faculty of Arts and Sciences) network, which in turn gets its connection through UIS (University Information Services), which provides networking for all of Harvard. Back in the good old days of 2000, before the file-sharing people went crazy, UIS had just upgraded its internet connection (to a 155MB/s OC-3, IIRC) -- oh what heady and naieve days those were. Now, however, the situation is this: file-sharing programs seem to act as a gas that consumes all available bandwidth. That first started happening, IIRC, the weekend when Napster was going to be shut down. Suddenly, the undergraduates doing file-sharing shut down the connection for the entire university (which is much larger than the undergrads).
This is a classic "tragedy of the commons" scenario. The file-sharing folks abused a shared resource and ruined it for everyone else. What the traffic shaping essentially does, since it limits the student network to some portion of the FAS-UIS feed, is allow the file-sharing programs to ruin only the undergraduate dorm commons. Now, it's easy to blame the shaping for the bad performance, but the real truth of the matter is that if the file-sharing programs weren't trying to consume essentially infinite traffic, your games wouldn't have a problem. The router doesn't slow things down just to mess with you. Gamers are getting "scrood" by the other undergraduates doing file sharing, not the traffic shaping. Incidentally, the shaping is similar to the sometimes-raised suggestion of giving the students their own internet connection and fighting it out amongst themselves, except that traffic shaping makes it easier for people to complain about "the Man" and that a separate feed would be stepping on UIS's toes a bit. Also, a student-only connection would have to be much bigger than an OC-3, because it has already been demonstrated that an OC-3 can't handle the file-sharing traffic.
What most people want is for the file-sharing people to be moved out of their commons and into someone else's. That's what you're suggesting when you want TCP but not UDP to be limited. As an off-campus user, the file-sharing people are already out of my commons, so I'm happy that I can access Harvard websites and mail and login servers again. Most of the users of Harvard's network aren't on-campus undergraduates, either. Perhaps you can understand, then, why I'm defending the shaping. I like the network to be actually usable instead of the packet-dropping mess that it is when shaping isn't there.
There's also the option of getting rid of the commons, which is the shaping-per-user suggestion, but that has some disadvantages, too. Even if the undergrad dorms get one third of that university-wide OC-3, that's only 7.5kb/s per undergrad, which is not too great. Up the per-user bandwidth to something reasonable and now a certain number of file-sharing people can take everything over again.
Then there's the get-a-bigger-commons option. There are several problems with this. First, it's not clear that there exists a pipe fat enough for the number of file-sharing users among the undergraduates plus the other uses from the university at large. Second, of course, is what someone else has already mentioned -- try to imagine the FAS network admins justifying to UIS the need for the university to get a bigger feed, and UIS in turn having to justify that budget item, just for undergraduate file-sharing.
Everything I have said is based on what limited stuff I have seen and heard, but it seems to me that it's all really complicated, and if there were an easy solution, I'm sure it would already have been adopted.
Re:Scarcity (Score:3, Funny)
Re:Proof (Score:2, Funny)
-Jason-