In beginning our BitTorrent project there are many issues to consider. First and foremost, the implications of creating a union between Apple’s iTunes network and service and a distribution source like BitTorrent. As it stands, the legal and ethical issues surrounding BitTorrent are vast and it seems risky, to say the least, for a respected and legitimate company like Apple to foster a business relationship with BitTorrent.
First one must examine simple the legality of the situation at hand, the biggest issue begin copyright infringement. BitTorrent’s notorious reputation as a peer-to-peer network used for sharing music and other data files poses a threat to the reputation of Apple and their already existing relationships with copyright holders. Secondly, the posed threat of security to a network using BitTorrent is severe. Because of the shared database driven nature of the program, Adware, Spyware and many other harmful programs risk being spread to users computers leaving them unhappy with their services – especially if they are paying for those services as with iTunes. In addition, one must consider the grave ethical issues posed with such a merger. On a network like BitTorrent, there is a lack of monitoring and control over the information and traffic. For example, pornography and other such undesirable forms of traffic and file sharing can easily occur under the nose of the administrators.
Before a company like Apple and BitTorrent could enter into any form of Business agreement or network merger, these issues and many more would have to be closely examined and BitTorrent would have to show serious dedication to the legitimacy of Apple’s business practices and adhere to more stringent legal and ethical protocols.
Monday, February 26, 2007
Sunday, February 25, 2007
Stephen Reading - Technical Liason: First Analysis of Costs of Peer to Peer Distribution
This post will use rough math to show that switching to Bit Torrent is a bad idea.
Let’s start with the myth of the long tail. “March data for the 1.1 million songs of Rhapsody, [a streaming music service], shows a 22% no-play rate; another 19% [of songs] got just one or two plays” [1]. This is 41% of songs that are almost never being streamed. Let’s assume that since Rhapsody charges a subscription fee but doesn’t require users to pay for each song they stream that users are more likely to stream a less popular song on Rhapsody to try it out than they are buy it on iTunes. Using that logic it’s probably fair to assume that 50% of songs get no more than two downloads each month on iTunes. Let’s also assume the ratios roughly correspond to Rhapsody with 25% getting no buys, 12.5% one buy, and 12.5% two buys. At 100 kB/s it will take about one minute to download each song.
365days/12months*24hours/day*60minutes/hour=43800 minutes per month.
2 minutes/43800 minutes/month = .0045 is the percent of time that 12.5 percent of the library is being downloaded
1 minute/43800 minutes/month = .0023 is the percent of time that another 12.5 percent of the library is being downloaded
And 0 is the percent of time 25% of the library is being downloaded.
Overall this is .0017 %.
This means that on average, for 50% of the iTunes catalog you have a .0017% chance of having anyone else also be downloading the file to give you any of the benefits of a peer to peer system. I won’t even get into the chance that they will have a good enough connection to help noticeably.
Another quote from the article makes the situation even bleaker. “Ecast told me that now, with a much bigger inventory than when Mr. Anderson spoke to them two years ago, the quarterly no-play rate has risen from 2% to 12%” [1]. This means that iTunes’ library of 3.5 million songs [2] probably has a much higher rate of songs with extremely low demand than we just assumed compared to Rhapsody’s library of 1.1 million.
Given how low that percent chance is I believe it is fair to guess (and it is just a guess because I couldn’t find data) that at least 75-80% of the music available would get no benefit from peer to peer because it is downloaded so infrequently. Indeed iTunes is probably more like Amazon where “2.7% of Amazon's titles produce a whopping 75% of its revenues” [1] which would put the percentage at 97.3 percent of music that would get no benefit from peer to peer delivery.
Let’s calculate Apple’s current bandwidth costs. Last year they sold just over 1 billion songs according to their reports. We will keep it at exactly 1 billion songs and balance it by counting all of the tv downloads as being last year even though it started in October 2005.
1 billion songs x 5 MB = 4,800TB
50 million TV shows x 350 MB = 17,000 TB
1.3 million movies x 1.25 GB = 1,600 TB [3]
Gives us an estimate of somewhere around 24000 TB of bandwidth used last year. At $.65 per gigabyte [4], we get 16 million dollars in bandwidth fees.
Let’s assume Apple is going to switch to Bit Torrent, what are their costs now? Bandwidth is $.45 per gigabyte and storage $.2 per gigabyte [4], and 75% of the library has to be downloaded entirely from Apple, but that 75% makes up only 25% of the total downloads. 24000 TB *1024 GB/TB *$.2 + 6000 TB *1024 GB/TB*$.45 = 7.7 million dollars. This means that if Apple gives no incentives to users to upload from their machine, and impossibly spends nothing on bandwidth seeding the files, they will save 8.3 million dollars. To put that into perspective, raising the cost of a song from $.99 to $1 would make them 10 million dollars a year. $8.3 million is also roughly equal to 3% of what they spent on advertising the iPod in 2005 [6] or their revenue in 2 and a half hours, if you prefer [7].
However if they want the scheme to work they should expect to give incentives to upload. Since bandwidth used to cost $.45 per gigabyte after storage fees that is the maximum we can expect Apple to pay. This means at best a user would have to upload over 450 songs before they got a single one free.
Now consider what rate a customer would be willing and able to upload at, given that a higher upload speed will limit their download speed and that incentivizing uploads with money will lead to a situation in which there are far more uploaders than downloaders, leaving no one to send data to even if it’s desired. Consider that for every person who gets a free song from uploading, we need 100 people to download a song without uploading a single bit. For convenience let’s say I’m on a decently fast connection that the program connects people to and I manage to upload at 12 KB/s. At that rate it will take me about 24 hours to upload 1 GB (how convenient!), earning me $.45 a day and taking me two and a third days to earn my first free song. At 130 watts to power my computer and about $.15 per kilowatt hour, it costs me about $.02 an hour to have my computer running [5], or $.48 a day. At that cost, Apple will not even pay me back what it costs me in my electric bill to upload for them.
As technical liason I think it is very clear that Apple should not switch to Bit Torrent anytime soon, unless they naively expect users to be willing to upload files for free. I will, however, be spending the rest of the project trying to think up ways to make this switch work, under the assumption that Apple is not worried about the fact that it won’t save them any money, will in fact cost them money to develop a new type of DRM that works with peer to peer networking, and is just generally a dumb idea.
[1] It May Be a Long Time Before the Long Tail Is Wagging the Web. Wall Street Journal reprinted at http://www.ntoddblog.org/connected/2006/07/index.html. Jul. 26, 2006.
[2] http://www.apple.com/itunes/store/
[3] http://www.apple.com/pr/library/2007/01/09itunes.html
[4] M. Ratcliffe. “YouTube wildly profitable?” http://blogs.zdnet.com/Ratcliffe/?p=186 Oct. 4, 2006.
[5] http://michaelbluejay.com/electricity/computers.html
[6] I. Fried. “Apple spends a bundle on iPod ads.” http://news.com.com/Apple+spends+a+bundle+on+iPod+ads/2100-1047_3-5978598.html Dec. 1, 2005.
[7] “One billion: Apple's first quarter profits soar”. http://www.thinksecret.com/news/0701q1earnings.html Jan. 17, 2007
Let’s start with the myth of the long tail. “March data for the 1.1 million songs of Rhapsody, [a streaming music service], shows a 22% no-play rate; another 19% [of songs] got just one or two plays” [1]. This is 41% of songs that are almost never being streamed. Let’s assume that since Rhapsody charges a subscription fee but doesn’t require users to pay for each song they stream that users are more likely to stream a less popular song on Rhapsody to try it out than they are buy it on iTunes. Using that logic it’s probably fair to assume that 50% of songs get no more than two downloads each month on iTunes. Let’s also assume the ratios roughly correspond to Rhapsody with 25% getting no buys, 12.5% one buy, and 12.5% two buys. At 100 kB/s it will take about one minute to download each song.
365days/12months*24hours/day*60minutes/hour=43800 minutes per month.
2 minutes/43800 minutes/month = .0045 is the percent of time that 12.5 percent of the library is being downloaded
1 minute/43800 minutes/month = .0023 is the percent of time that another 12.5 percent of the library is being downloaded
And 0 is the percent of time 25% of the library is being downloaded.
Overall this is .0017 %.
This means that on average, for 50% of the iTunes catalog you have a .0017% chance of having anyone else also be downloading the file to give you any of the benefits of a peer to peer system. I won’t even get into the chance that they will have a good enough connection to help noticeably.
Another quote from the article makes the situation even bleaker. “Ecast told me that now, with a much bigger inventory than when Mr. Anderson spoke to them two years ago, the quarterly no-play rate has risen from 2% to 12%” [1]. This means that iTunes’ library of 3.5 million songs [2] probably has a much higher rate of songs with extremely low demand than we just assumed compared to Rhapsody’s library of 1.1 million.
Given how low that percent chance is I believe it is fair to guess (and it is just a guess because I couldn’t find data) that at least 75-80% of the music available would get no benefit from peer to peer because it is downloaded so infrequently. Indeed iTunes is probably more like Amazon where “2.7% of Amazon's titles produce a whopping 75% of its revenues” [1] which would put the percentage at 97.3 percent of music that would get no benefit from peer to peer delivery.
Let’s calculate Apple’s current bandwidth costs. Last year they sold just over 1 billion songs according to their reports. We will keep it at exactly 1 billion songs and balance it by counting all of the tv downloads as being last year even though it started in October 2005.
1 billion songs x 5 MB = 4,800TB
50 million TV shows x 350 MB = 17,000 TB
1.3 million movies x 1.25 GB = 1,600 TB [3]
Gives us an estimate of somewhere around 24000 TB of bandwidth used last year. At $.65 per gigabyte [4], we get 16 million dollars in bandwidth fees.
Let’s assume Apple is going to switch to Bit Torrent, what are their costs now? Bandwidth is $.45 per gigabyte and storage $.2 per gigabyte [4], and 75% of the library has to be downloaded entirely from Apple, but that 75% makes up only 25% of the total downloads. 24000 TB *1024 GB/TB *$.2 + 6000 TB *1024 GB/TB*$.45 = 7.7 million dollars. This means that if Apple gives no incentives to users to upload from their machine, and impossibly spends nothing on bandwidth seeding the files, they will save 8.3 million dollars. To put that into perspective, raising the cost of a song from $.99 to $1 would make them 10 million dollars a year. $8.3 million is also roughly equal to 3% of what they spent on advertising the iPod in 2005 [6] or their revenue in 2 and a half hours, if you prefer [7].
However if they want the scheme to work they should expect to give incentives to upload. Since bandwidth used to cost $.45 per gigabyte after storage fees that is the maximum we can expect Apple to pay. This means at best a user would have to upload over 450 songs before they got a single one free.
Now consider what rate a customer would be willing and able to upload at, given that a higher upload speed will limit their download speed and that incentivizing uploads with money will lead to a situation in which there are far more uploaders than downloaders, leaving no one to send data to even if it’s desired. Consider that for every person who gets a free song from uploading, we need 100 people to download a song without uploading a single bit. For convenience let’s say I’m on a decently fast connection that the program connects people to and I manage to upload at 12 KB/s. At that rate it will take me about 24 hours to upload 1 GB (how convenient!), earning me $.45 a day and taking me two and a third days to earn my first free song. At 130 watts to power my computer and about $.15 per kilowatt hour, it costs me about $.02 an hour to have my computer running [5], or $.48 a day. At that cost, Apple will not even pay me back what it costs me in my electric bill to upload for them.
As technical liason I think it is very clear that Apple should not switch to Bit Torrent anytime soon, unless they naively expect users to be willing to upload files for free. I will, however, be spending the rest of the project trying to think up ways to make this switch work, under the assumption that Apple is not worried about the fact that it won’t save them any money, will in fact cost them money to develop a new type of DRM that works with peer to peer networking, and is just generally a dumb idea.
[1] It May Be a Long Time Before the Long Tail Is Wagging the Web. Wall Street Journal reprinted at http://www.ntoddblog.org/connected/2006/07/index.html. Jul. 26, 2006.
[2] http://www.apple.com/itunes/store/
[3] http://www.apple.com/pr/library/2007/01/09itunes.html
[4] M. Ratcliffe. “YouTube wildly profitable?” http://blogs.zdnet.com/Ratcliffe/?p=186 Oct. 4, 2006.
[5] http://michaelbluejay.com/electricity/computers.html
[6] I. Fried. “Apple spends a bundle on iPod ads.” http://news.com.com/Apple+spends+a+bundle+on+iPod+ads/2100-1047_3-5978598.html Dec. 1, 2005.
[7] “One billion: Apple's first quarter profits soar”. http://www.thinksecret.com/news/0701q1earnings.html Jan. 17, 2007
Stephen Reading - Technical Liason: Progress So Far - Investigation of Competition and Similar Projects
Democracy Player
Allows content creators to publish channels traditionally by HTTP or by Bit Torrent.
To start I downloaded NBC Nightly News from Feb 24, 2007 on Sunday Feb. 25, 2007 at 2:50 pm. Average Speed = 286 k/s, Top Speed = 410 k/s, Total Download Time = 6m:40s. Length 21:27, size 112 MB, Video Rate 89 kB/s.
The quality level was significantly worse than the standard 350 MB/hr (142 kB/s) rip that I believe is necessary for this technology to catch on. At this speed it would take 20 minutes to download that quality of file.
Next I downloaded from the Terra channel : Signs Of Life: The Search for Life on Mars: Part One. Downloaded Sunday Feb 25, 2007 at 3:00 pm, size 94.1 MB, length 11:57, download time: 5m:20s, average speed: 301 kB/s, video rate 129 kB/s. The quality level was noticeably worse than the standard tv rip, but watchable.
Itunes Video Store
Current implementation of the product we are analyzing.
Video comes at 640x480 resolution, which is good, but doesn't go full screen ever as far as I can tell from the previews. At $2 per episode regardless of whether it is half hour or hour long show the pricing scheme feels as if it is ripping off the customer for shorter shows instead of giving them a deal for longer shows. I could not determine the file size of tv shows.
Movies are 1 to 1.5 GB. At 1.25 GB even at a high speed of 300 k/s it would take 1.2 hours to download. Currently since it is not BitTorrent the file is downloaded sequentially which means you can watch the movie before it is done downloading. This means (assuming constant download speed) that as long as the total time it takes to download the file is less than the time it takes to watch it, you can begin watching immediately upon purchase. If Apple switches to BitTorrent the entire file will have to download before it can be watched, which will most likely frustrate the user more and more as file size increases. This means video quality will have to be balanced against user impatience as Apple evaluates the potential of HD content. Since there is currently not much way to watch the video on a tv instead of a computer screen, increased quality would seem unnecessary given the smaller screens.
Sopcast
Uses peer to peer network to stream live tv from a limited number of channels.
OK quality (320x240), but frequent pauses for buffering due to lack of users. Received about 100 kB/s download speed when trying to watch Barcelona vs Athens on GOL TV on Sunday Feb. 25, 2007. 54 other fans were watching the channel at the time.
Since it is p2p live tv buffering works differently than normal web video. It is possible for the buffer percentage to decrease even while you download if you don't download fast enough. This is because a node that is behind all the other computers isn't able to upload useful data. So people with slow connections will see nothing, instead of taking longer to see something as might be expected.
From experience, during the World Cup there were often 1000 people watching, which didn’t improve quality but allowed for no pauses for buffering. CCTV 5 (China Central Television Sports) is usually the one with the most viewers.
Joost
Streaming TV distributed through a peer to peer network,f rom the creators of Skype and Kazaa.
I signed up for Joost Beta weeks ago but have not been accepted yet, and invitations to join seem to be rare.
Bit Torrent TV Rips (eztv for this example)
Illegal rips allow for download of full episodes of copyrighted tv shows with no drm.
According to the site the typical hour long Bit Torrent TV rip is 608x336 resolution at a compression rate of 142 kB/s. The rules for these communities stipulate that files must be under 350 MB/hr.
From this and other file sharing sites it appears that:
1080p movies = 20 GB
1 hr (aka 40 min) 1080i tv shows are ~4.2 GB
1 hr (aka 40 min) 720p tv shows are 1 GB
So called HR releases are 1 hr (aka 40 min) 700 MB at an unknown resolution since I can’t download it.
According to HDTV Info Port these analog measures convert into the following digital resolutions:
720p - The picture is 1280x720 - 60 complete frames per second.
1080i - The picture is 1920x1080 - (60/2 interlaced frames per second)
= 30 complete frames per second.
1080p - The picture is 1920x1080 - 60 complete frames per second.
Even the lowest quality format that qualifies as HDTV, 720p, has three times more pixels than Apple’s 640x480, which would mean tripling file size and download time.
Allows content creators to publish channels traditionally by HTTP or by Bit Torrent.
To start I downloaded NBC Nightly News from Feb 24, 2007 on Sunday Feb. 25, 2007 at 2:50 pm. Average Speed = 286 k/s, Top Speed = 410 k/s, Total Download Time = 6m:40s. Length 21:27, size 112 MB, Video Rate 89 kB/s.
The quality level was significantly worse than the standard 350 MB/hr (142 kB/s) rip that I believe is necessary for this technology to catch on. At this speed it would take 20 minutes to download that quality of file.
Next I downloaded from the Terra channel : Signs Of Life: The Search for Life on Mars: Part One. Downloaded Sunday Feb 25, 2007 at 3:00 pm, size 94.1 MB, length 11:57, download time: 5m:20s, average speed: 301 kB/s, video rate 129 kB/s. The quality level was noticeably worse than the standard tv rip, but watchable.
Itunes Video Store
Current implementation of the product we are analyzing.
Video comes at 640x480 resolution, which is good, but doesn't go full screen ever as far as I can tell from the previews. At $2 per episode regardless of whether it is half hour or hour long show the pricing scheme feels as if it is ripping off the customer for shorter shows instead of giving them a deal for longer shows. I could not determine the file size of tv shows.
Movies are 1 to 1.5 GB. At 1.25 GB even at a high speed of 300 k/s it would take 1.2 hours to download. Currently since it is not BitTorrent the file is downloaded sequentially which means you can watch the movie before it is done downloading. This means (assuming constant download speed) that as long as the total time it takes to download the file is less than the time it takes to watch it, you can begin watching immediately upon purchase. If Apple switches to BitTorrent the entire file will have to download before it can be watched, which will most likely frustrate the user more and more as file size increases. This means video quality will have to be balanced against user impatience as Apple evaluates the potential of HD content. Since there is currently not much way to watch the video on a tv instead of a computer screen, increased quality would seem unnecessary given the smaller screens.
Sopcast
Uses peer to peer network to stream live tv from a limited number of channels.
OK quality (320x240), but frequent pauses for buffering due to lack of users. Received about 100 kB/s download speed when trying to watch Barcelona vs Athens on GOL TV on Sunday Feb. 25, 2007. 54 other fans were watching the channel at the time.
Since it is p2p live tv buffering works differently than normal web video. It is possible for the buffer percentage to decrease even while you download if you don't download fast enough. This is because a node that is behind all the other computers isn't able to upload useful data. So people with slow connections will see nothing, instead of taking longer to see something as might be expected.
From experience, during the World Cup there were often 1000 people watching, which didn’t improve quality but allowed for no pauses for buffering. CCTV 5 (China Central Television Sports) is usually the one with the most viewers.
Joost
Streaming TV distributed through a peer to peer network,f rom the creators of Skype and Kazaa.
I signed up for Joost Beta weeks ago but have not been accepted yet, and invitations to join seem to be rare.
Bit Torrent TV Rips (eztv for this example)
Illegal rips allow for download of full episodes of copyrighted tv shows with no drm.
According to the site the typical hour long Bit Torrent TV rip is 608x336 resolution at a compression rate of 142 kB/s. The rules for these communities stipulate that files must be under 350 MB/hr.
From this and other file sharing sites it appears that:
1080p movies = 20 GB
1 hr (aka 40 min) 1080i tv shows are ~4.2 GB
1 hr (aka 40 min) 720p tv shows are 1 GB
So called HR releases are 1 hr (aka 40 min) 700 MB at an unknown resolution since I can’t download it.
According to HDTV Info Port these analog measures convert into the following digital resolutions:
720p - The picture is 1280x720 - 60 complete frames per second.
1080i - The picture is 1920x1080 - (60/2 interlaced frames per second)
= 30 complete frames per second.
1080p - The picture is 1920x1080 - 60 complete frames per second.
Even the lowest quality format that qualifies as HDTV, 720p, has three times more pixels than Apple’s 640x480, which would mean tripling file size and download time.
Wednesday, February 21, 2007
Subscribe to:
Posts (Atom)