Home > Archive >  2009 >  August >  19

How to fix URL-shorteners

Wednesday, August 19, 2009 by Dave Winer.

First a few notes as a preamble. 

1. URL-shorteners are bad for the Internet. They centralize linking, and make it more fragile, and more controllable. Wait till the Chinese govt finds out about them.  

2. When bit.ly breaks, it will be an outage that may be bigger than Twitter going down. Not only do we lose the present, but we lose the past too. One big URL shortener that dominates the others is itself a dangerous thing. 

3. Twitter could and should obviate the need for URL-shorteners. Yes I know SMS messages are limited to 160 chars. So shorten the URLs at the SMS gateway and leave them long for communication over pathways that are not so limited. Any engineer could see this obvious solution.  

4. For now URL-shorteners are a fact of life. 

End of preamble. Now to what is needed in URL-shorteners to work around the various issues they present. 

It's not so different from the problem with Feedburner, and the solution they used, and implemented quickly once it was known. 

1. CNAMEs. It must be possible for the user to own and control the domain his or her URLs live at. Technically, this means I register the domain name, and map a sub-domain to the URL-shortener site with a CNAME record. Anyone who knows how to use Godaddy can do it. I would be happy to write a howto that explains.  

2. Shared data. The URL-shortener and the user share a space where the data is stored. Joe Moreno at Adjix, who I have been working with, has figured out how to do it on Amazon S3. I have mapped a domain to an S3 bucket, and given his software permission to write to that bucket. Here's the key point. At any time I can revoke the permission and my URLs still work. Or Adjix could disappear, and the shortened URLs would still work. With this method the only way there is linkrot is if S3 goes down.  

Here's a URL that links to a Flickr picture: 

http://tmp.loose.ly/jmxe  

Obviously the sub-domain, tmp.loose.ly, is temporary. But if you're a techie, I encourage you to do a DNS lookup on tmp.loose.ly. You'll see it's a CNAME to s3.amazonaws.com. And get the contents of the file to see how it works. It's static. Yet it still gathers statistics. Yes, it's unusual. That's why Joe was the only one to crack this nut. He's a creative guy.  

It's such a clean implementation that if I decide later to move the files to an Apache server on Linux, no problem.  

I think basically Adjix has solved all the problems with URL-shorteners. I hope other engineers poke at this and verify my conclusion or disprove them.