Bandwidth Demand at American Universities

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."

Comments

Its not the bandwidth everywhere.

[LWolenczak]

I attend a UNC system school, and we have NCREN powering the internet access. A little background info, everybody has one or two ds3 or oc3 lines. NCREN has an oc48, an oc12, and a handfull of oc3 lines uplinking them to the world. The problem is, that NCREN has given oc3/ds3 lines to companies like microsoft, and they load down the entire system. The traffic graphs for my university never really exceed 40mbit/sec on one line, and 12mbit/sec on the second line. We even have a third line, and its dark most of he time.

But the NCREN oc48, oc12, and all the oc3 line outbound are allways loaded at 99%. In ncren before you hit the core routers, pings are below 30 ms, once packets leave, they have gone as high as 2000 ms. NCREN refuses to admit there is a problem, or resolve the issues. One problem also seems that one of sprints core routers that peers with NCREN is faulty or over loaded.

I must admit, when I lived in the dorms, it was nice to sometimes be able to play the major ra3 servers at 35-55ms.

Morpheus is the killer

[baptiste]

Sure, legitimate bandwidth use is high. But student use of bandwidth is huge.

The university I work at has a huge pipe (1Gb I think) shared with two other local universities. Generally we use the least amount of the bandwidth, but at one point our usage had hit like 500Mb/s Needless to say teh other schools were freaking - they were losing packets due to teh pipe being so full. Well, our dorms are on their own network. Sure enough, thats where most of the bandwidth was going. Blocking Kazza/Morpheus and co is tough since it'll switch and seek out other ports. So the only solution was to limit the total bandwidth for the dorms to 25Mb/s Sure enough, once that block went in place our usage overall dropped to like 90Mb/s. 300-400Mb/s of bandwidth just for the dorms????

The students were upset since their pipe was now slam full and they had trouble getting out, but the response basically was - stop running servers and stuff for music that suck up bandwidth and you'll be able to get on the Net to do the stuff you need to do. Its not perfect, but for now it works and keeps us from totally saturating our pipe.

90% of the people = 10% of the bandwidth use

[reddawnman]

I work in the Academic Computing Services department of one of the UC Schools...

The majority of our problems come from about 10% of the population on campus who are online 24/7 downloading their pr0n, thus giving most of the other students probably polled in this study a reason to start asking for more bandwidth. I should also add that these are the same 10% who are hogging internal bandwidth playing counterstrike, etc.

I think that the term "Insatiable Demand" is definitely a misnomer. Although the "Prominence of file sharing" does apply to quite a few people in our dorms, 90% of the people are utilizing the network for, at most 10-20 megs a day. In fact, we have a 2Mbit cap on the routers coming out of the dorms, and most users find that they can surf the web and get their 3 or 4 files a day with no problems, and are pleased that, at 4AM, they can get an insanely high throughput. The reason that the students complain about the network being slow is because of the caps (which most don't know about) at peak times, because, again, the 10% that actually do have an unquenchable thirst for data would take full advantage of the situation.

I should add also that we block Morpheus, thereby removing those oh-so-lovely TCP standards hacks it implements, so YMMV

Re:90% of the people = 10% of the bandwidth use

[davew2040]

Always good to sneak in a jab at computer games and the people who play them in a post like this. However, since games like Counterstrike typically only consume some 1.5kB/s-3kB/s to operate normally, what exactly is meant by "hogging internal bandwidth"? Those numbers seem like very normal use of campus lines, even if the total packet throughput is above-average in a realtime UDP application.

Re:90% of the people = 10% of the bandwidth use

[AnotherBlackHat]

Yeah, I know what you mean. I used to help with the administration of the pianos in my college, and it was much worse. Seems like the music majors consumed way over 90% of the time available, even though the other students barely touched them. More than once I was on the receiving end of complaints that there weren't enough pianos. And I heard that the swimming pool was in a worse situation, with the swim team hogging nearly 95% of that valuable resource.

The worst part was the huge number of administrators who seemed to think that the situation was completely acceptable. I would constantly hear phrases like "The equable distribution of resources does not mean everyone must use exactly equal portions.".

Each of us is an individual. And the name of that individual is Clancy Jones. - Clancy Jones #148.

Re:The same thing happens in the UK

[sully67]

Clarification:

ja.net charges for transatlantic bandwidth partly because a great deal of ja.net's costs were on paying for transatlantic lines. Anything that doesn't travel via these lines (ie via LINX or GEANT is free which means most resources in the UK or europe.

Ja.net has also mitigated the need to use so much transatlantic traffic through the use of mirror.ac.uk and the National ja.net webcache.
If only more people were to use these, a number of smaller UK academic institutions are not aware of a number of services that ja.net can provide for them but this is changing in recent years.

The billing costs eat you alive

[Animats]

Telephony billing costs more for off-peak periods than the transmission costs. Billing, marketing, collection, and support for an ISP cost much more than the back-side bandwidth. It's not economic to meter. And it kills you competitively.

You need some throttling to hold back the few percent of the user population who will suck up all available bandwidth, but the practical cap is maybe 10x the median.

As a technical matter, I hate packet drop as a throtting measure. Packet reordering is much more effective at throttling TCP, especially for long TCP connections.

[I used to do network congestion research. I invented "fair queueing", discovered "congestion collapse", and was the first to describe the "tragedy of the commons" problem for networks. I introduced all three of those phrases to networking. See the RFCs that bear my name. So I do know something about this, although I got out of networking and into graphics years ago. J. Nagle]

Different things occurred in New Zealand

[AtomicBomb]

downloading linux isos from mirror.ac.uk
I am sure I will get my butt severely kicked by my net admin if I tried to download a Linux iso from outside. Mirroring is the king here.

Due to the very small pipe that we've got (we are amongst the bottom of internet bandwidth scale throughout major universities in Asia-Pacific according to the now dysfunction Asian weekly survey last year), we cannot afford to download anything big. We are in CSE/EEE. In a dept with about 700 person, only 2-3 staff member are authorised to download something as big as an iso. All the others need to use the internal mirror.

Student needs to pay from their own account for using internet (NZ$0.4/MB). Staff and PhD students has "unlimited" internet access (ie, you will get cut off if *monthly* download > 100MB). It is much less than ideal. But, we somehow survive. ;-)