a modest proposal

As you've probably noticed, even stranger things than usual are happening to the update schedule. . . Life is hard. You've heard it before. We wrote this article for fun, because it combines our otaku natures and techie inclinations. . . Now, of course, someone needs to implement it. ^_^

In some ways, serving anime is much like serving other large quantites of data, except of course that it is impossible to make any sort of profit. You can't win. Most days you can't even break even. but hey, we're otaku.

By the way, minor boring technical detail ahead. Not too much.

1. location

Repeat after me 'oh what a friend i have in .edu'. Find a grad student who'll let you use his office as a colo. Put it in the dorms. Find some closet with a network port that no one uses. Point out to the administrators how convenient it'll be for students on-campus, not having to tie up incoming bandwidth.

Try not to make life too hard for whomever's hosting it. It'll already be making weird 'clunk' noises throughout the day, spinning up, loading, unloading, etc.

2. hardware

The most essential aspect of this server is its storage space. Conventional solutions involving large hard drives are both expensive and error-prone. The solution, as with so many low-budget computer problems, lies with yesteryear's hardware.

Consider, for example, HP's surestore line of optical jukeboxes. We found one for $100, cast off by some branch of California's labyrinthine government. It holds 238 double-sided 5.2 GB MO media for slightly over 1.2 TB of storage, more than enough for the average club. (And yes, it was fully loaded.)

Although the jukebox has only four drives, it's easy to add a basic caching implementation, simply by reading the entire requested file into a hard drive buffer. Note that there will be a delay between the initial request and the file's actually being sent. This should be under a minute, however. Depending on the hardware, it may be under ten seconds. In a DCC send, this is acceptable.

The costs are:

MO array - 100
miscellaneous - 50

That's it. (This assumes, of course, that your time is worthless, that you have old computers lying around, and that you can convince your local university to pay power and bandwidth. ^_^ have fun.)

Computing hardware is not an issue. The optimum solution would be an old workstation. These machines tend to feature integrated SCSI, slim, low-powered designs, and excellent thermal engineering. The example that we have to hand is a sparcstation 10.

3. software

Stick to something remotely manageable. IRC isn't infallible. Power may go out. People might upload enough to fill the buffer space (hah). Ideally, physical access to the box will be required no more than once a month. In practical terms, that means a UNIX variant. Linux jukebox support is a bit iffy. (Read: virtually nonexistent.) Try solaris or hp-ux if possible. (Our sparcstation will take solaris 8 just fine.)

IIRC, the jukebox driver doesn't actually cache the files on its own. The software allows for each side of each disk to be mounted on a seperate path, mediating access transparently. The server, whether IRC or FTP, will need to call a routine that copies the file into the buffer, and call another routine to remove the file when downloading is complete. (Trivial to implement.)

IRC is the preferred protocol, since it has vastly more sophisticated queue management. FTP is widely-supported, but can easily be monopolized by a relative handful of users. Although IRC lacks user authentication, we could equivalently create a closed channel and regulate access that way. IRC also allows for easy advertising, although this sort of server wouldn't really need to advertise its presence.

In general, uptime is not an issue. People tend to be understanding, particularly since most fserves are run off home computers, behind home connections. The standards are not that high.

4. miscellany

look! good fortune is all around you!

*shrug* We weren't able to actually buy the autoloader. It would have been nice, but we didn't start thinking about it until later. A lot of the ideas we present haven't been tested, and shouldn't be implemented without considerably more thought by sane people.

This is just an example. The general idea is broadly applicable, though, to whatever hardware you happen to find. It's rooted in the idea that, online, we survive as otaku within the interstices. We live on cast-off technology, what we can buy and what the mainstream will let us take.

But we share. Somehow we've managed, by and large, to play well with others.

—anonymous coward ( heavily edited by c-chan )