Craig Labovitz of Arbor Networks put out this conspiracy theory that there was wide spread effort to “manipulate P2P traffic” using some flawed statistical analysis.  The story made it to the front page of Slashdot yesterday as some sort of evidence of wide scale global sabotage of peer-to-peer (P2P) applications.  But the analysis relied on some very questionable use of statistics and methodology.

The first sign that something is wrong is the use of very liberal use of percentages and arbitrary Y scale boundaries that were designed to exaggerate minor statistical differences.  The second problem is all of the charts shown relied solely on percentage scales rather than absolute numbers.  I replicated two of the charts.

Based on these two relative traffic level charts that Lebovitz produced, Lebovitz posited:

“As a side note, the cyclical inverted traffic pattern of P2P is interesting in its own accord. The inversion is highly suggestive of either persistent congestion or, more likely, evidence of widespread provider manipulation of P2P traffic rates.”

But this is silly because we don’t know the absolute traffic levels.  For all we know based on what little data Lebovitz presented, P2P traffic might have even gone up in the middle of the day even as it drops in percent traffic.  To illustrate my point, I took Lebovitz’s percent data and produced a hypothetic traffic level chart where P2P traffic actually increased during the middle of the day.  So if web, P2P, and other traffic did look like the chart below, the percent traffic levels would actually correspond with Lebovitz’s charts.  That means it is possible that P2P traffic level during the day isn’t dropping and Lebovitz’s charts don’t really tell us anything because it’s short on data.

Even if P2P traffic did drop during the day, that still makes sense because P2P users often run their P2P in the background while they sleep in the night so that they don’t have to put up with severely degraded broadband performance when they’re trying to surf the web and do other interactive things.

Every Internet user who plays online games or uses Voice over IP (VoIP) knows that P2P is toxic to their application because P2P applications cause massive latency and jitter.  Trying to surf the web with a P2P application going full blast is like going back to the days of dial-up Internet service because the P2P application can consume 90% of the broadband connection leaving the web browser and everything else only 10% of the bandwidth pie.  This is because P2P can cheat the Internet’s normal congestion control mechanism by opening up hundreds of connections (what network engineers call “flow”) at the same time.  The Internet’s congestion control mechanism tries to give every flow the same amount of bandwidth rather than giving every user on the network their fair share.  So if there is P2P traffic shaping going on so that bandwidth is distributed fairly, that would be a good thing!

UPDATE 9/3/2009 – Craig Labovitz updated his analysis with data showing P2P usage compared to itself over the course of the day.  It does show P2P usage dropping, but it still doesn’t prove manipulation by the network operators.  Everyone I know turn off their P2P clients during the day when they want to do other things with their broadband connection.  If I have P2P running, I know for sure that my VoIP calls and game play will be completely unusable.  If I need to surf the web, I’d be sure to at least throttle my own P2P speed down in half.  Even if ISPs are managing their networks, manipulation carries negative connotations and implies wrong doing.  If the network operators are closing the multi-flow loophole that P2P exploits in TCP congestion control, that is just sensible network management.