If geeks love it, we’re on it

Will BitTorrent’s switch to UDP kill the internet?

Will BitTorrent’s switch to UDP kill the internet?

It is said that P2P traffic accounts for nearly half of the bandwidth consumed on the Internet. It is believed by some that this usage is a fraction of what is possible were it not for the ease in which TCP traffic is managed for the sake of QoS. With the announcement that makers of BitTorrent intend to use UDP — a relatively unmanaged counterpart — some believe that core infrastructure of the Internet will be crushed once this traffic escapes local ISPs.

Transmission Control Protocol, better known as TCP, is used to deliver and receive over 98% of all traffic generated on the Internet today. It is such a critical component of the Internet Protocol (IP) suite that Internet traffic is simply referred to as “TCP/IP” traffic.

TCP has become such a star because it ensures reliable and orderly delivery of information. It is responsible for initiating, negotiating and maintaining a connection to virtually any server with the guarantee that information won’t be corrupted when it is sent or received. This error correction and avoidance comes at the expense of speed, as parties participating in the exchange of traffic must yield bandwidth to the overhead of reliable delivery.

Meanwhile, TCP’s counterpart, the User Datagram Protocol (UDP), another member of the IP suite, comprises less than 2% of all traffic generated today. It has fallen into relative disfavor because it guarantees neither the delivery nor integrity of the data it is charged with managing. While it has become the transport — that is, sending and receiving — protocol of choice for delay-sensitive services like IPTV, VoIP and online gaming, its unreliability has made it a poor choice for web serving, email, or other bulk data delivery services.

Because of UDP’s disuse, a vast majority of the focus invested in managing Internet traffic has been done on TCP. Because of TCP’s structure and design, it can be easily filtered, safely delayed, and easily throttled to provide end users a quality service. Even without the intervention of politics, agendas or economy, TCP dismisses speed for control. For these very reasons, the makers of BitTorrent have decided to abandon TCP in favor of UDP.

BitTorrent, the company behind the amazingly popular meshing P2P protocol, has released the newest version of their uTorrent client with strict UDP use in tow. While ISPs, like Comcast, that use bandwidth throttling will remain unaffected, core Internet servers are innocently designed to be traffic-agnostic. More succinctly, what may be fifty percent of all Internet traffic may soon switch to a protocol that cannot be orderly controlled as in the case of TCP.

In this case it is easier to view the alleged conundrum from the perspective of real-world road traffic. Under a TCP system, all traffic stays within its lanes and obeys the rules of the road per traffic lights and signs. Though congestion may occur, these drivers safely and reliably arrive at their destination every time.

This is in opposition to a UDP system that can best be described as a screaming Autobahn where drivers may or may not arrive all by their wholesome. These roads are an uninterrupted stretch of white-knuckle speed. Much like UDP-based applications must perform the error checking that TCP would otherwise perform, drivers in a bender are ferried off to the hospital with the hope that they’ll pull through.

In point of fact, BitTorrent is an ideal UDP application. Through a hashing algorithm integral to the BitTorrent service, the protocol has always been able to assure that it has reliably received all pieces of the information from all the users providing them. Anyone who has stopped and restarted a Torrent has observed the software’s progress in verifying the integrity of the received data against the master template provided in the .torrent file. But does BitTorrent’s new affair with UDP spell Internet disaster?

Bennett is correct in stating that ISPs will probably not be affected. Many, like Comcast, are now switching to protocol-agnostic bandwidth throttling which will artificially narrow the pipe to heavy users. Bennett claims that after this traffic escapes local regulation, it will hurtle along the unregulated autobahn and the whole system will come to a grinding halt as core servers struggle to manage an unmanageable transport protocol.

In truth, I’m conflicted about this matter. While proponents of bandwidth management have religiously beat the “fifty percent of internet traffic!” drum, there’s no concrete evidence that this number is correct. Similarly, it is hard to determine how much of this traffic consists of that which is generated by BitTorrent.

While core infrastructure managers have tools at their disposal to weather Bennett’s alleged storm front, none of them are politically correct. One such technology, Deep Packet Inspection, selectively discards packets by analyzing their contents for protocols that have landed on a black list. To Comcast’s chagrin, the FCC did not take kindly to its use on the enormous slice of the Internet that is US soil. Alternatively, core infrastructure could simply reject any and all UDP packets, but doing so would wipe IPTV, VoIP and gaming off the map.

Bennett makes a hard argument in favor of arming ISPs with the tools to prevent uTorrent from “killing the internet.” He explicitly calls for the abolition of laws that stand in the way of this goal, and tacitly implies that new ones should be created to ensure this goal. Yet in the Internet’s practical history of 18 years, the wait and see approach has proved reliable and satisfactory.

If anything, general history has proven that preemptive strikes often prove overzealous. Again, Comcast’s decision to roll with DPI serves central to proving this point. But more than that, the Internet has been a shining example of the old adage that says “if it ain’t broke, don’t fix it.” Nothing is broken yet, Mr. Bennett, and I’d rather we wait to attack the potential problem with data and evidence rather than suppose and conjecture our way into the unforgiving embrace of network non-neutrality.

Comments

  1. kryyst
    kryyst I for one welcome my new UDP masters.

    But seriously I see this as both good and bad. In terms of getting me my torrents faster it's potentially a good thing. But at the same time if it starts to actually impact VoIP, gaming and other technologies -Cisco VPN over UDP, for example. Then it's a problem. I mean right now my torrents aren't actually coming in slow....
  2. Thrax
    Thrax Right. It could potentially be bad, but many things on the internet are potentially bad. Seems like there's one new THIS WILL KILL THE INTERNET AS WE KNOW IT post every week. Wait and see. Wait and see.

    I, too, welcome our new UDP overlords.
  3. Kyle Agreed that there should be a 'take a breath' moment here and actually figure out what is happening today and what could potentially happen tomorrow. It will be interesting however to watch what impact this might have on ISPs use of DPI, potentially this time in a more pragmatic (and most would argue, ethical) manner.

    I welcome no overlord, but looking forward to watching them closely.

    <a href="http://www.breakingpointlabs.com&quot; rel="nofollow">Kyle</a>
  4. Linc
    Linc To put in grade-school terms: ISPs started it.
  5. CB
    CB
    Keebler wrote:
    To put in grade-school terms: ISPs started it.
    qft.

    It's like the nerdy kid that everyone's been picking on, finally lashes out, and he's the one that gets detention.
  6. shwaip
    shwaip
    CB wrote:
    qft.

    It's like the nerdy kid that everyone's been picking on, finally lashes out, and he's the one that gets detention.

    Hey. I was under 18, the records are sealed. You're not supposed to know about the incident.
  7. drasnor
    drasnor Anywhere else this would be called growth. Here's an idea; ISPs should use some of that money we send them every month to upgrade their infrastructure if they're starting to get strained rather than sinking cash in R&D for "congestion management". I fork over extra cash for the phat pipe so they'd better guarantee the service I'm paying for.

    -drasnor :fold:
  8. Richard Bennett You might enjoy my follow-up:

    http://www.theregister.co.uk/2008/12/05/richard_bennett_bittorrent_udp/

    A lot of questions remain about the soundness of the BitTorrent approach.

    Richard Bennett
  9. Thrax
    Thrax I actually read your article this morning and was glad to see that you had posted a followup. I am in complete agreement that BitTorrent owes anyone interested in the health of internet traffic complete details regarding the underpinnings of uTP.

    My support for uTP is apparent, but I am no more interested in the implosion of the A-level internet than anyone else. I hope that the protocol's development under the eyes of the IETF will put my confidence on solid ground.

    In the end, I guess we'll see how things pan out! Thank you for taking the time to comment on my piece, Richard. :) It's good to hear from you.
  10. Khaos
    Khaos I believe that this is a good step for the evolution of the Internet. Our over-reliance on TCP is not healthy for long-term development of the network. Particularly in media applications, such as streaming, high-definition audio/video, the possibility of moving most of the data validation from the packet level to the chunk level via post-processing of traffic on local machines shows a lot of promise.

    I can imagine this software technology being adapted to provide streaming HD A/V on a short delay to allow for buffering and hash verification of data chunks.

    By the way, I think that the reaction to this news by some people is completely overblown. While it is true that UDP traffic is not as well routed as TCP traffic, that much we can fix, and in the long term, transferring large quantities of *known* data will be far more efficient over UDP. With UDP, you don't have the overhead of ACK packets constantly going back to the server! Like I said, very promising stuff. It might take the infrastructure a little while to catch up with the increase in UDP traffic, but once it does, it will be a good thing moving forward.

Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!