Sunday 3 February 2008

BitTorrent Traffic Vs the rest of Internet Traffic

I read with horror how a chief technical officer at APConnections publish an article with misinformation (i guess the aim is to convince users not to use BitTorrent).

Here is what I disagree; [My comments in square brackets like this]

First of all, lets look behind the scenes at how BitTorrent can affect a large ISP's network.

Torrents create thousands of connections. If you use a BitTorrent program on your laptop and attempt to download content (music, movies, et cetera), your computer will locate and find as many sources of this specific content as possible and download from all sources simultaneously. [true]

Conceptually, this is like simultaneously downloading separate chapters in a book from different sources, and then putting the chapters back together on your computer when the downloads complete. Your motivation is time; downloading all the chapters simultaneously is much faster than downloading them in sequence. [good conceptualisation]

In addition to looking for multiple sources for a single file, BitTorrent clients allow the user to attempt more than one file download at a time. [that's the beauty of BT!] This behavior further multiplies the number of active connections from your computer to the Internet. The amount of concurrent connections generated by one home user to the Internet can reach into the thousands. [Untrue! Most torrent client sets the TOTAL connections to around 400] Compare that number of connections with somebody who passively browses the web, and perhaps runs an AIM messenger client on their home computer; their connection total would not exceed fifteen or so at any moment in time.

But multiple connections generate tremendous overhead for a provider, too. [FALSE]

Each connection from your computer across the Internet must be accounted for [???] in various switching points (routers) in order for the data to flow. [OK, data packets go through] You can imagine these switching points as little traffic police standing at the intersection of a street during a 5K road race telling the runners where to turn. Now imagine thousands of races running all over the city at one time; each starting and terminating at a different location. Obviously the traffic cops would get overwhelmed and have to restore some order, like making sure organizers did not run their races all at the same time. [Bad conceptualisation. The effectiveness of Internet traffic is based on packet switching. That means that each packet may traffic from source to destination on different paths. The little traffic police at the switch does not care whether this packet is for the same connection to the last packet. Each packet is routed on path depending on the current traffic situation. CONNECTIONS do NOT increase any workload for the traffic police. The amount of traffic DOES!] In a nutshell, that is one of the motivations for a service provider to try and limit torrents. [I don't think the motivation is because of number of connections. Rather it is more related to the AMOUNT of traffic] The spontaneous and overwhelming number of connections they can create can overwhelm their circuits.

Where do all these BitTorrent files come from? Who is serving up this content?

When you set up a BitTorrent client and start downloading files you also unwittingly become a BitTorrent host/server. Any file you download to your computer can also be sent to other users while your BitTorrent program is active. [True]

From a content provider perspective there is great value in having your data hosted by millions of home users connected by BitTorrent; they can reduce your distribution costs. A content provider that distributes content from servers hosted on their private servers would be charged accordingly for the amount of data uploaded from their hosts. By using a BitTorrent model for distribution, however, a content provider becomes a parasite on third party sites allowing them to replicate and distribute content like a benevolent virus. [Bad comparison and deliberately biased against bitTorrent! Are you saying that you are objecting your users to host home-base smaill web servers. They are serving content as well. When a user connects to a website, the initiation is also out-going traffic! Again, a packet is a packet and is a packet. To the ISP, except for accounting purposes, incoming or outgoing packets should be treated the same and have similar load on routers.]

So who is absorbing the distribution cost when a content provider distributes their data by releasing it into the BitTorrent cloud? [Good questions, but it should be asked from an accounting point of view rather than technical.]

The additional cost of delivering BitTorrent content is essentially shouldered by the major ISPs who likely did not account for this traffic overhead when building out their networks. [No!, ISP do NOT have another additional cost except users are actually requesting the ISP to meet the sale agreement. For unlimited quota plan, most ISP will offer different prices based on different speed limits. So users using BT only INCREASE the amount of traffic, which is agreed in the price. For limited quota plan, BT users are godsend. They need more traffic and you have a great reason to ask them to upgrade! Another point to note, once a physical link is built (eg an optical fibre) the cost of running the fibre is small compared with the capital investment. Amount of traffic does NOT have any effect on running cost! Quota is a mechanism in accounting!]


If I am wrong in my interpretation of Internet technology, I stand to be corrected. Please cite sources when pointing out my mistake.

1 comment:

Albert Ip said...

PS I believe ISP usually oversell their capacity. It was OK when most users were less than 100% uploading or downloading at peak traffic/speed. When every user is at near peak traffic, the infrastructure is unable to cope.

Also most ISP will run proxy servers to minimize backbone traffic. When users are downloading content from large publishers, the proxy servers are running at their peak efficiency. However, if users are downloading via BitTorrent, proxy servers are basically useless. BT traffic then translates to backbone traffic (where as fetching content from large publishers are not.)