So, as has already been reported elsewhere on the Internets, Comcast (of which I am a current customer) is going to assess a 250 GB limitation on transfer for its residential customers. This is more than a little disturbing, being that I signed up for unlimited use Internet.

Disclaimer: The math in this post is quick, back-of-a-paper math. It may contain errors; if so, please let me know. Also, bear in mind that I am not talking about net data transfers here, I am also including protocol overhead for TCP/IP and PPP where relevant, which is part of the reason why the numbers will seem so large in a few situations. Feel free to verify the math, though.

I decided to take a look at what kind of transfers I have effected since the last time my router was rebooted. Here’s the raw data (I am pretty sure that the numbers reported by ifconfig count things like retries and the like as well as protocol overhead for TCP/IP, but I am not 100% positive on that—I’ll go ahead and make that assumption for the moment):

Saturday, 2008-Aug-30 at 13:55:01 - mbt@zest - Linux v2.6.24
Ubuntu Hardy:[0-16/698-0]:~> ssh allspice 'uptime && /sbin/ifconfig inet0|grep "RX bytes"'
 13:58:34 up 44 days,  4:52,  5 users,  load average: 0.00, 0.00, 0.00
          RX bytes:91387801756 (85.1 GB)  TX bytes:51854749824 (48.2 GB)

This shows that my box has been up for 63,652 minutes, and has transferred in that time a whopping 143,242,551,580 bytes in that time. This is an average of 2,250,401.43 bytes (2,197.66 KiB, 2.15 MiB) per minute that the connection is up. For a 30-day month with no leap seconds or minutes, this means that in 30 days, the average transfer would be somewhere in the area of 97,217,341,611.51 bytes (94938810.17 KiB, 92,713.68 MiB, 90.54 GiB). In other words, about 36.21% of the threshold that Comcast is talking about. And this has actually been a rather light period of Internet usage for us—I have taken no snapshots of large projects on the Internet during this period, nor have I mirrored the Hardy repositories yet (I like to keep a local mirror on my server, and keep it up-to-date, when I can, so that I can install new systems without running out for updates), nor have I downloaded any other large data this month. This is just regular, every-day usage for us.

Out of all of that, most of the transfer is probably mine. I sit in several IRC channels all the time, I occasionally play Guild Wars (and even that, well, that’s not factored into this, because I’ve only played it once and I haven’t downloaded the entire world yet, because I haven’t been into it for a long time and didn’t have a local copy when I installed it again), I pull source code for packages when I am working on them, and so forth. I use a lot of bandwidth, and it’s not for commercial purposes.

Now, something else to think about before going further: Comcast claims that the limit will be 250 GB, not 250 GiB. In other words, 250×1000^3, not 250×1024^3, which is the series of measurements that I use. Converting my 97,217,341,611.51 estimated monthly usage, this results in a figure of 97.21 GB out of an allowed 250 GB (38.89%, not 36.21% as computed above). This assumes that they’re being accurate in the first place. Since I can’t be 100% sure either way, being that there is so much confusion of computer size measurements out there these days, I’ll assume that they’re using powers of 10 instead of powers of 2 for size computation, since this means that I will err on the side of safety.

So, I estimate that we chew through approximately 40% of their quota per month. What do I run on my home network? I have an Asterisk server, which routes all of my calls to multiple extensions (everyone should have their own PBX, it makes life really easy!), which uses VoIP. It doesn’t see that much in the way of usage, so it’s a small player. I do a lot of file transfers between the Internet and here, both over IPv4 and IPv6. Some of the overhead in our network usage is actually that from using an IPv6 tunnel to Hurricane Electric, though I am not sure just how much overhead there is compared to IPv4 directly over the existing Internet infrastructure.

These numbers are pretty awesome to me, given that once upon a time, I had a link to the Internet that was easily measured in bits per second.

In any event, if Comcast wants to limit everyone residentially to 250 GB per month of transfers, why don’t they have their accounts set that way? The account that I have currently has 8 Mbps downstream and 1 Mbps upstream; 9Mbps total. Assuming a 30-day, zero-leap minute/second month (IOW, each day is 86,400 seconds, and there are 30 of those, for a total of 2,592,000 seconds per month), this means that I can transfer up to 23,328,000 Mbits in a single month—which works out to 186,624,000 MB, or 182,250 GB. In order to bring that number down to 250 GB per month, I would have to be limited to transferring 32,000 Mbit per month, or 0.0123456790 Mbps aggregate in both directions—a whopping total of 12.64 Kbps. Yes, you could transfer more data than that on a dial-up connection in a one month span. In comparison, a half-duplex 28,800 kbps modem can transfer up to 37324800000 Kbits, 37324800 Mbits, or about 4,665.6 GB per month, assuming steady, synchronous, error-free data transfer and a damn good phone line.

In effect, Comcast wants for its users to have less impact on its network than a dialup user would have, to say nothing of 38,400 kbps modems or better. Now, the reality of a 28,800 kbps modem is slightly different, being that connections over phone lines tend to suck, and so there are errors introduced in the bitstream between the modems, which lowers the effective bitrate of the connection while retries are made at the link layer. However, to bring that 4,665.6 GB down to 250 GB, 94.64% would be the approximate rate of error—way more than I think anyone would tolerate unless they absolutely had to.

Now, that having been said, the real data transfer is always less. These numbers are assuming that the underlying protocols’ overhead (e.g., TCP/IP on Ethernet, PPP and TCP/IP on dialup) are factored into the data transfer rates. But, those would be considered to be measured as part of the link anyway, since that is data that is being transferred. I am not saying that you can download a 9,331 MB file on a dialup connection in a month, just that that much data would be transferred if the link were constantly saturated.

Is this a reasonable limitation? Hell, no it’s not.

We’re moving into times where everything takes place online. Computers aren’t really used as standalone machines anymore—a lot of online communication is tied to services, and those services communicate a lot of data. Email, Usenet, and IM are all technologies which permit people to communicate with each other, but for some time now, we’ve had systems that communicate with each other, as well. if you want the weather, you get that from an online service. If you want news, you can get that from an online service. Presence notification, and other things, are not things that humans directly communicate to each other; these are types of communications that are computer-to-computer, on a human’s behalf. When I go away, everyone gets a notification of that event when they are signed on to an instant messaging service that I use and they have me on their roster/buddy list. That’s a sizeable chunk of data if a lot of people are watching you.

Add to this the fact that we watch video streams online—and video streams are not lightweight things—and listen to things like Internet Radio, and other types of data transfer which weren’t possible even 15 years ago on a wide scale, and you can begin to see the problem. I don’t listen to Internet Radio all the time; I only listen about 10 hours a week. The stream is 128Kbps MP3, and for one 128Kbps MP3 I have which is 5 minutes 56 seconds, the file size is 5,707,152 bytes. MP3 is a frame-based format, and is easily streamed from any point in the file, making it not dependent on having to read a header. This particular file has no metadata in it, so it should be a decent sample to use. The file size if it were a 10 hour file, then, would be in the ballpark of 577,127,730 bytes, or 0.577 GB. For a month which averages 4¼ weeks, that means 2.446 GB just for the stream. Make that stream last the entire month, and it will be 41,553,196,584 bytes, or 41.553 GB. Add in TCP/IP overhead, which is 40 bytes for every 1460 bytes in the stream (and there are 28,461,094 such chunks in a month-long MP3) and you have 1,138,443,742 bytes of protocol overhead, for a total of approximately 42,691,640,326 bytes transferred for each person that listens to an Internet Radio stream for the entire month. 3 of us makes that potentially 128,074,920,978 bytes, about 128.07 GB. With a 250 GB cap, that’s over 50% of it, right there. Add that to our 97.21 GB figure from earlier, and that’s a total of 225.28 GB. Damn close to the limit. And if Comcast were to use “creative math” in determining the figures, that could be over the limit. Depends on what they include, if they include data overhead from the CPE, and so forth that I cannot possibly begin to account for.

In short, 250 GB is NOT a reasonable limit. Even less so, if you are a household with more than one person in it. So, what do I have to do? Program my routing, NATting server so that it notifies me when/if we do get close, because it is entirely within the realm of possibility, and force us to be disconnected if we get too close to the limit. They should cut our bill by 75% for this, and even that wouldn’t be enough—250 GB of transfer, in 2008? That’s just cheap.