The old adage that “a lie told often enough becomes true” seems to have withstood the test of time for good reason: it actually seems to work.  Despite the fact that the BitTorrent protocol continues to be one of the least network friendly applications in the world based on my test results of the protocol, BitTorrent continues to tell the world that their application is “network friendly”.  BitTorrent’s VP Simon Morris replied to me via email that they’ve made improvements to their protocol, but acknowledged that my test results were within their design expectations.  The problem is that even with the newly improved BitTorrent protocol, they are still one of the worst offenders in terms of the tendency to cause other applications problems.  Yet all these claims of BitTorrent’s network friendliness are merely repeated and amplified by the blogosphere and other advocacy groups as fact when none of them actually do any testing of the protocol or cite any specific research to backup their claims.

The latest version of uTorrent (the most popular BitTorrent client software in the world which BitTorrent acquired in 2006) employs an integrated speed test user interface that can poll various servers from dslreport or M-Lab).  The results could then be used by uTorrent to automatically set up speed limits to optimize file transfer throughput, but potentially at the expense of causing even more congestion and network jitter.  Previously, users had to manually measure their own speed and be smart enough to limit the upload speed to roughly 90% of their peak upload capacity to get optimum performance.  In my tests when I allowed uploads to consume nearly 100% of the available capacity, it dramatically reduced download speeds (probably due to the degraded ability to send receive acknowledgements).  With the addition of automated speed testing, uTorrent will be able to hit maximum download speeds out of the box and therefore cause maximum congestion in the network.  Figure 1 and 2 show speed test results from uTorrent, speedtest.net, and pingtest.net.

Figure 1 – uTorrent speed test results

Figure 2 – speedtest.net and pingtest.net results

So while these features to optimize the throughput of uTorrent, BitTorrent hasn’t bothered implementing any meaningful network friendly features.  For example, they have yet to implement a low latency inducing mode that would allow you to continue using uTorrent without worrying about severely degrading your VoIP or online gaming application with jitter.

BitTorrent decides to be selfish rather than friendly

In 2008, I proposed a more friendly BitTorrent protocol that would space out packets evenly to reduce jitter in the BitTorrent protocol.  Simon Morris of BitTorrent recently emailed that they took my suggestions seriously and even researched the issue soon after I made my proposal.  Morris stated that the engineering was difficult, but their biggest reluctance stemmed from the fact that making BitTorrent space packets out evenly had too much of a negative impact on BitTorrent throughput.  In other words, BitTorrent made the conscious decision to be selfish and impolite to grab as much bandwidth as possible.  Being polite like VoIP and online gaming applications which send out evenly spaced packets to minimize jitter would have meant ceding network capacity to impolite protocols.

Now perhaps the rationale for being selfish rather than polite may very well be business driven since hardcore BitTorrent users want their files to transfer as quickly as possible regardless of the impact to other protocols.  BitTorrent’s founder Bram Cohen developed BitTorrent to use up as much bandwidth as possible and he doesn’t care if the ISPs don’t like it.   But it is misleading to claim that BitTorrent is “network friendly.  Furthermore, this mindset is short sighted because it ignores the fact that many BitTorrent users (or their family members) are also VoIP users and online gamers who are forced to shut down BitTorrent during the day out of concern for BitTorrent congestion.

The best solution is to solve these problems in the network

There is a better solution that would fix these problems at the network level where applications like BitTorrent could be allowed to be as aggressive as possible and get as much bandwidth as possible while protecting jitter sensitive applications like VoIP or online gaming.  Let’s say that a broadband connection has 8 megabits per second (Mbps) of downstream capacity and 1 Mbps of upstream capacity.  If we prioritized VoIP and online gaming applications and dynamically “carved out” up to 0.15 Mbps of upstream and downstream capacity, we could let other applications grab at least 7.85 Mbps downstream and 0.85 Mbps upstream capacity.  If VoIP or gaming isn’t in use, the other applications like uTorrent can grab all of the bandwidth.

With this kind of intelligent network management in place, BitTorrent would be able to perform at its peak throughput with no negative impact on VoIP or gaming applications.  That means BitTorrent users (most of whom are also VoIP and/or gamers) don’t need to turn off BitTorrent during the day to avoid harming the other applications they use.  The net result would be more active BitTorrent users adding to the available pool of “seeders” who make content available to other BitTorrent users and that actually improves BitTorrent performance for everyone on the Internet.  The network based solution is also superior to users manually throttling their BitTorrent clients to only 25% of peak capacity to reduce (but not eliminate) the damage on other applications.  On an 8 Mbps broadband service, the network based solution allows users to download at 98% performance for BitTorrent with no damage to VoIP or online gaming.

Irrational opposition to network based solutions

This type of solution is what I referred to as “true neutrality” where low bandwidth jitter sensitive applications are protected from other aggressive applications.  It is obviously the best solution to the problem where all applications get the most performance possible and coexist in harmony.  Yet the hard line Net Neutrality advocates like Free Press and Public Knowledge want us to buy into their religion of the dumb “First In First Out” (FIFO) networks.

BitTorrent also opposes network based solutions and Simon Morris insisted that only the application developer should solve these problems and indicated that we don’t know if network operators can be trusted.  Yet BitTorrent has demonstrated that they have no desire to be more network friendly and they like the law of the jungle where the most aggressive applications like BitTorrent grab as much bandwidth as possible at the expense of other applications.  They merely want to be perceived as “network friendly” even if nothing could be further from the truth and they want regulators to ban broadband providers from good network management practices under the banner of “Net Neutrality”.  They won’t do anything to protect VoIP and gaming at the application level but they don’t want the network to solve these problems.

Fortunately, we do have some smart people like Julius Knapp at the FCC that understand the need to protect applications which are sensitive to jitter caused by other applications.  It gives me some faith that the FCC will fix their proposed regulations to allow for reasonable discrimination.