Eponymous, there's another problem also:
"The problem is that we are bandwidth limited to the rest of the world,
and connection limited overall. The low rate we're getting to the rest
of the world (15 Mb/s or less most of the time), and the server limit
of 1500 active connections combine to make connections difficult regardless
of where you connect from. All 1500 connections are continuously in use
pumping out data a low rate to the rest of the world leaving no local
high-speed connections available.
Eric J. Korpela"
In other words, all high-ping-slow-modem-users that manages to get a connection is blocking out the gigabit-internet2-connectors.
But I'll guess there's a much higher probability to get a connection via internet2, and then you get one, you'll only tie up the connection for 1 second.
Looking on the graphs, it's obvious seti was down for some hours yesterday.
After something was changed at the packetstrangler last evening, there seems the campus-connection is back to normal. The last couple of days, it was 5-10 mbit below normal...
Today at around 12 it looks like they'we done some more changes, resulting in 5 mbits more.
Oh, and I've come across another comment that doesn't bode well for the future:
"Unfortunately, throwing more bandwidth at the problem only postpones the problem, whether or not P2P clients are dealt with.
At last week's NAG meeting, Cliff Frost showed us some graphs based on network usage over the past three months, and they were revealing.
The summary:
* P2P clients consume 20-30% of the 70Mbps pipe from the main campus. Total campus traffic, including the dorms, is more like 50 P2P, but much of it is under the dorms' separate 40Mbps cap. (Dorm traffic is something like 80% P2P, but the students pay for it with their own fees, and to some extent they are free to decide to do that).
* Last November, http traffic was in the 300-350 gigabytes per day range. Now, it's 400-450 gigabytes per day. kazaa+gnutella combine to be 150-250 gigabytes per day (they are spikier than http traffic). (I may have the units wrong, I'm doing this from memory). The other big bandwidth usages on campus are SETI@Home, NNTP, and FTP. The first two aren't as important, because they both are configured to use mostly excess bandwidth. FTP is small compared to HTTP; it looked like about half of kazaa.
What this means is, killing all the kazaa/gnutella users, even if it were possible, wouldn't solve the problem for long. In six months, our http traffic will have swelled by 150-250 gigabytes per day--the trend line is obvious. And what percentage of the http traffic is any more appropriate than kazaa/gnutella?
There are rarely technological solutions to social problems.
Let's say we do try to place significant technical restrictions on kazaa/gnutella traffic. Maybe we use the plan suggested here of "whitelisting" http, ssh, and ftp traffic, and sectioning off everything else. The first thing that happens is we spend a lot of time and effort troubleshooting things, like IPSEC, or Corporate Time, that we missed on the initial whitelist. Then, some kid comes up with the bright idea of using port 80 for gnutella, and then what do we do? And even if everything works perfectly, 6 months down the road we'll be hitting the cap with whitelisted traffic alone.
I think eventually the campus will be forced to start charging users for bandwidth usage over a certain baseline. There are provisions for that in the network funding model which spawned the node bank. CNS is now capable of tracking traffic down to a single IP, but they need quite a bit more data before they can start billing down to a single IP; there are a lot of thorny issues around that. It's really similar to phone billing, but people aren't used to the idea of network billing, and it will take a while before usage habits change.
peer-to-peer filesharing is a good example of the tragedy of the commons; a collection of individuals, each acting in his own best interests, creates a situation that is negative for everyone involved. (Other examples are the morning commute and the parking situation near campus, which also lack technical solutions, and which most propsed remedies actually make worse).
The original "The Tragedy of the Commons" article by Garrett Hardin in Science is available at . The example of a communal pastureland for grazing is instructive to our case.
--
Tom Holub (tom_holub@LS.Berkeley.EDU, 510-642-9069)
College of Letters & Science
249 Campbell Hall"
Facebook
Twitter