Apparently CacheLogic based most of their conclusions on data collected from a European "tier 1" ISP. However, another study by Sandvine found regional differences in file sharing networks. Europe and the US don't have the same file sharing patterns, or even popular file sharing programs. http://www.sandvine.com/solutions/pdfs/Euro_Filesharing_DiffUnique.pdf Of course, there is always CAIDA's data. Peer-to-peer analysis is on their long range plans. http://www.caida.org/projects/progplan/progplan03.xml
On Thu, 15 Jul 2004, Sean Donelan wrote:
Apparently CacheLogic based most of their conclusions on data collected from a European "tier 1" ISP. However, another study by Sandvine found regional differences in file sharing networks. Europe and the US don't have the same file sharing patterns, or even popular file sharing programs.
I would also like to add that over here Direct Connect is quite common among the more organized and hard-core file swappers, while the really-hardcore guys of course are still using private ftp servers. With proliferation of 10 meg ethernet (full duplex) connections for residential use in (especially) northern europe and in asia, users are more likely to serve content to other users around the world. I have made some studies regarding the bandwidth usage pattern between equal size populations where the difference is if they have ADSL 8M/800k or if they have 10M/10M. The amount of data served is 1/3rd on ADSL compared to the symmetric ethernet population, and as a population they serve out more content than they download (approx twice the amount) on ethernet. The ADSL population peak at approx twice the bw as they serve, but on average they serve a little less than they download. Hmm, the above wasn't very clear, but here it goes in another format: Ethernet: Peak almost twice upload as download. Average is 2.5-3 times more upload than download. ADSL 8M/800k: Peak twice the amount download as upload Average is 1.3-1.5 more download than upload Upload bw usage is almost flat over time Download bw peak is approx double the average level. My interpretation of this is that p2p networks are quite intelligent in using the available bandwidth, and that Copyright holders only solution is a "content crunch" due to providers limiting their users upload potential due to heavy usage, such as capping the amount of bandwidth allowed per month or alike. -- Mikael Abrahamsson email: swmike@swm.pp.se
On Jul 15, 2004, at 5:25 AM, Mikael Abrahamsson wrote:
On Thu, 15 Jul 2004, Sean Donelan wrote:
Apparently CacheLogic based most of their conclusions on data collected from a European "tier 1" ISP. However, another study by Sandvine found regional differences in file sharing networks. Europe and the US don't have the same file sharing patterns, or even popular file sharing programs.
I would also like to add that over here Direct Connect is quite common among the more organized and hard-core file swappers, while the really-hardcore guys of course are still using private ftp servers.
With proliferation of 10 meg ethernet (full duplex) connections for residential use in (especially) northern europe and in asia, users are more likely to serve content to other users around the world.
I have made some studies regarding the bandwidth usage pattern between equal size populations where the difference is if they have ADSL 8M/800k or if they have 10M/10M. The amount of data served is 1/3rd on ADSL compared to the symmetric ethernet population, and as a population they serve out more content than they download (approx twice the amount) on ethernet. The ADSL population peak at approx twice the bw as they serve, but on average they serve a little less than they download.
Hmm, the above wasn't very clear, but here it goes in another format:
Ethernet:
Peak almost twice upload as download. Average is 2.5-3 times more upload than download.
ADSL 8M/800k:
Peak twice the amount download as upload Average is 1.3-1.5 more download than upload Upload bw usage is almost flat over time Download bw peak is approx double the average level.
My interpretation of this is that p2p networks are quite intelligent in using the available bandwidth, and that Copyright holders only solution is a "content crunch" due to providers limiting their users upload potential due to heavy usage, such as capping the amount of bandwidth allowed per month or alike.
Let's hope that their users don't try to do things like videoconferencing from home. (Like I do.)
-- Mikael Abrahamsson email: swmike@swm.pp.se
Regards Marshall Eubanks T.M. Eubanks e-mail : marshall.eubanks@telesuite.com http://www.telesuite.com
On Thu, 15 Jul 2004, Marshall Eubanks wrote:
Let's hope that their users don't try to do things like videoconferencing from home. (Like I do.)
Have you calculated the amount of BW you use with your video conferencing? The usage of savvy p2p-using households can be in the hundreds of gigabytes/month which I doubt you'll get with your videoconferencing? 100 kilobits/s over a month is approx 35 gigabytes. -- Mikael Abrahamsson email: swmike@swm.pp.se
Sean Donelan wrote:
Apparently CacheLogic based most of their conclusions on data collected from a European "tier 1" ISP. However, another study by Sandvine found regional differences in file sharing networks. Europe and the US don't have the same file sharing patterns, or even popular file sharing programs.
http://www.sandvine.com/solutions/pdfs/Euro_Filesharing_DiffUnique.pdf
If you leave BitTorrent out, which is probably the fastest growing protocol out there, the statistics are missing about one third of the bits moved. Pete
On Thu, Jul 15 05:27:32 2004 From: Mikael Abrahamsson Ethernet: Peak almost twice upload as download. Average is 2.5-3 times more upload than download. ADSL 8M/800k: Peak twice the amount download as upload Average is 1.3-1.5 more download than upload Upload bw usage is almost flat over time Download bw peak is approx double the average level. My interpretation of this is that p2p networks are quite intelligent in using the available bandwidth, and that Copyright holders only solution is a "content crunch" due to providers limiting their users upload potential due to heavy usage, such as capping the amount of bandwidth allowed per month or alike. mikael thanks for the hard data. YAY! but that strikes me as a generous interpretation of it. your data is also consistent with the fact that ethernet-attached universities and other campuses tend to have more lenient AUPs, as well as higher bandwidth, than do DSL/cable networks. it might be the humans that are intelligent, not the p2p networks. (forgive me for having more faith in system administrators than p2p software engineers :) ) also our grad student thomas studying p2p traffic tells me that there is no sense of localization in most (if not all) p2p networks; so i am more likely to download a movie from an (advertised as) ethernet user in Asia than from downloading from an (advertised as) DSL user next door. and my understanding it that's all based on how the user configures their p2p servent, it's not like the network figures out the available bandwidth to potential remote clients. that's a pretty loose definition of intelligence, believing everything you hear from an end host :) [i'd be happy to be wrong about the above, btw...] wrt capping the amount of bandwidth per month, that strikes me as a [yet another] stall not a solution. it may be a traffic engineering solution. it doesn't even approach being a copyright holder solution. but i reckon copyright solutions belong on another list. per seandonelan's reference, yes caida will have 2 papers on p2p traffic analysis by the end of the year, titles 'is p2p traffic dying or just hiding?' [take a guess] and 'transport layer identification of p2p traffic'. check caida web site in october. but neither study touches on the regional differences, i suspect it's a function of the relative popularity, ease-of-use, and type-of-content-served of certain applications in each place. i lend less weight to the differences than the similarities. killer app, indeed. k
It is all very interesting. Why we did not have such research reported on last NANOG meeting?
also our grad student thomas studying p2p traffic tells me that there is no sense of localization in most (if not all) p2p networks; so i am more likely to download a movie from an
Interesting. Are there any p2P systems which optimize traffic by localizyng it, when possible?
(advertised as) ethernet user in Asia than from downloading from an (advertised as) DSL user next door. and my understanding it that's all based on how the user configures their p2p servent, it's not like the network figures out the available bandwidth to potential remote clients. that's a pretty loose definition of intelligence, believing everything you hear from an end host :)
per seandonelan's reference, yes caida will have 2 papers on p2p traffic analysis by the end of the year, titles 'is p2p traffic dying or just hiding?' [take a guess] and 'transport layer identification of p2p traffic'. check caida web site in october.
but neither study touches on the regional differences, i suspect it's a function of the relative popularity, ease-of-use, and type-of-content-served of certain applications in each place. i lend less weight to the differences than the similarities. killer app, indeed.
k
Alexei Roudnev wrote:
Interesting. Are there any p2P systems which optimize traffic by localizyng it, when possible?
Most p2p applications keep the connections which provide data at better speed and drop the ones which don´t. The effectiveness of this criteria varies from application to application. As far as I´m aware there are no applications that look at the locality of the data. There are many different approaches to increase the locality by modifying the traffic in various ways, from simple ones like ours which allows you to mark packets or announce prefixes over BGP4 for your network to police the p2p heavy remote-and-expensive prefixes to more extensive ones where the actual packet contents are modified to steer the payload traffic. Depending on the aggressiveness of the applications, this works differently. DirectConnect and Bittorrent are very aggressive on adapting their network configuration while eDonkey or FastTrack seem to be more relaxed. Obviously content availability also plays a big factor here since it´s hard to download from somewhere which does not yet have the bits. As far as I know no p2p networks do larger scale topology calculations but all base their activity on local selfish host behavior. Which is probably right on the mark 90% of the time anyway. Pete
On Sat, 17 Jul 2004, k claffy wrote:
servent, it's not like the network figures out the available bandwidth to potential remote clients. that's a pretty loose definition of intelligence, believing everything you hear from an end host :)
Well, if the p2p program gets chunks of data from all sources, anyone with a faster uplink will complete faster and thus get more chunks (as the p2p program requests new chunks from each peer as the previous one is finished), and thus produce more traffic. Not much intelligence, just the way they work. -- Mikael Abrahamsson email: swmike@swm.pp.se
On Sun, 18 Jul 2004, Mikael Abrahamsson wrote:
Well, if the p2p program gets chunks of data from all sources, anyone with a faster uplink will complete faster and thus get more chunks (as the p2p program requests new chunks from each peer as the previous one is finished), and thus produce more traffic. Not much intelligence, just the way they work.
http://www.itic.ca/DIC/global/2003/09/Top_ISPs_by_P2P_Activity.jpg
On Sun, 18 Jul 2004, Sean Donelan wrote:
http://www.itic.ca/DIC/global/2003/09/Top_ISPs_by_P2P_Activity.jpg
"Average ratio of active P2P nodes / Available IPs". That doesn't have much to do with number of bytes transferred, right? http://www.itic.ca/DIC/global/2003/09/FastTrack_Servers_Location_by_Country.... This is more interesting. SE consists of 9 million people, of which approximately 150-200k households have ethernet connections, most via Bredbandsbolaget. Then again, these numbers are not the amount of bytes transferred again, but the number of nodes. I guess Fasttrack is popular here in .se anyway. -- Mikael Abrahamsson email: swmike@swm.pp.se
participants (6)
-
Alexei Roudnev -
k claffy -
Marshall Eubanks -
Mikael Abrahamsson -
Petri Helenius -
Sean Donelan