http://in.tech.yahoo.com/041103/137/2ho4i.html
According to Reuters, BT is more traffic than web/other forms of traffic? I'm thinking the sampling methodology here might be a little skewed.
Then again, I could be biased. Any other facts that would support this?
DJ
1) where was the measurement done?
2) how was the measurement done?
3) what population was sampled?
On some networks BT might account for far more than 30%, on others far,
far less... Perhaps the writers will answer?
-Chris
For those not familiar, BitTorrent is a file sharing app that is commonly
used for exchanging full movies. As such, folks are moving gigabyte
files regularly and it's not surprising that this is detectable.
Shuffling .mp3's around would be trivial by comparison.
Tony
Reality check
This week's netflow for the Internet 2
http://netflow.internet2.edu/weekly/20041025/
has BitTorrent taking up about 4.8 % of the traffic, http is 15 to 18%, and all
file sharing is about 10%, down from 50% 2 years ago.
Since file sharing and related uses are generally heavy traffic sources on I2, I would conclude
that the Reuter's numbers are too high.
regards
Marshall Eubanks
It's also used for distributing large patches (XP SP2), the latest
ISO's of various (free) operating systems, any pretty much anything else
that would create a flash crowd load on a system that it could not handle
without distributing the traffic amoung the people downloading.
This isn't a made-for-pirating-software/audio program, don't treat it as
such.
Marshall Eubanks wrote:
Reality check
This week's netflow for the Internet 2
Netflow is based on port numbers and many run bittorrent on fairly random ports. Look at the 30%+ unidentified on the report.
Pete
Yes, but HTTP tends to run on the same port, and it only made 15.76% of bits and 18.53% of the packets.
I know P2P is "big", but is HTTP really only 16% of the bits on the 'Net?
Question is: Is this data representative of "Internet 1"? I'm thinking not, since "Iperf" was more bits than HTTP.
Hello;
> Netflow is based on port numbers and many run bittorrent on fairly
> random ports. Look at the 30%+ unidentified on the report.Yes, but HTTP tends to run on the same port, and it only made 15.76% of
bits and 18.53% of the packets.I know P2P is "big", but is HTTP really only 16% of the bits on the
'Net?Question is: Is this data representative of "Internet 1"? I'm thinking
not, since "Iperf" was more bits than HTTP.
The fascinating thing with the time history of these data is that Napster used to
dominate. Then it was killed and the "unknown" category steadily grew to replace it.
The rise of Iperf is recent and seems silly. What, of course, is not clear is what fraction
of capacity it represents - maybe it is a small faction of what could be used.
I thought that BitTorrent (due to its shared use of bandwidth) would use well known ports, but
if not then it is clearly part of the "unknown." One wonders how Reuters and company could measure
it, but here is a back of the envelope guess.
BitTorrent is just under 50% of the _known_ P2P traffic. Assume that it is also 50%
of the _unknown_ P2P traffic. That gives it a known fraction of the total traffic of
4.94 % (measured) and 15% (guesstimated), or about 20%, which is larger than http.
So, its plausible that BT traffic is > http traffic, but I wouldn't want to further than that.
Yes, I would assume that P2P is a substantial fraction of I1 traffic. It certainly goes on at work.
Regards
Marshall
Not really,
Most popular bittorrent websites force you to use ports other than 6881.
So netflow reports are inaccurate.
My guess is that you could account a large chunk of 31.59% Unidentified
to bittorrent.
Regards,
Bas
Yes, but, netflow (in terms of ip src/dst, protocol type, port numbers) is a
poor way of classifying traffic that works in a fashion similar to what
we're discussing here. P2p file sharing protocols is one instantiation..
SIP is another.
The commercial tools for classifying deeper than header info a la netflow
are out there already, although they may not be as "slick to deploy" as
netflow (which brings its own challenges) by turning on knobs in software on
existing routing equipment. The limitation is how motivated is the business
to deploy the gear, not whether viable equipment exists for exactly that
purpose..
Regards,
Christian
And which kind of gear is this, except for boxes which need a mirror of
the complete traffic ?
There are a couple of NetFlow (9/IPFIX) meters that can additionally to
the normal fields can also pass eg the first 100 bytes of a conversation
in either direction. But those are also meters on a seperate box and not
embedded in the hardware of routers (or is there anything I missed 
Greets,
Jeroen