CoralCDN

» see coralcdn.org as it lived, on the wayback machine

~ the eulogy ~

The entire user interface was a suffix. You took a URL that was not loading, put .nyud.net on the end of its hostname, and asked for that instead. A decentralized DNS layer read the new name and worked out which cache sat closest to you, and that cache handed you the page. In the early years you had to add the port too, so people were pasting links that ended in .nyud.net:8090 into forum threads without ever finding out what was on the other side.

What it solved was the oldest failure mode on the web. A small site gets linked from a large site, the traffic arrives in one lump, the server falls over or the hosting bill goes vertical. CoralCDN fixed that for free and without the site owner's participation, because anyone reading the thread could coralize the link themselves. The homepage put it as a slogan, first "You supply the content; we'll supply the scalability" and later the better one, "You make it fun; we'll make it run".

It was research. Freedman published the Coral design with Eric Freudenthal and David Mazières in 2004, built on a sloppy distributed hash table arranged in rings by network distance so that lookups stayed local. In March 2004 it went live on PlanetLab, the shared academic testbed running on a few hundred machines donated by universities. By December that year it was carrying amateur video of the Indian Ocean earthquake, which is not a load anyone had budgeted for.

This was infrastructure for the English-language technical web, and its only visible surface was a strange-looking link in somebody's comment thread. I cannot tell you I ever coralized anything, and the honest reason is that you never had to know it existed for it to work on you. Somebody upstream pasted the longer link and the page loaded.

~ the dossier ~

~ the trick ~

The 2010 retrospective Freedman wrote after five years of running it is unusually honest about what an open proxy network is actually like to operate. Anyone could point it at anything, which meant it inherited every abuse problem an open relay has ever had, and the original plan of opening it to volunteer machines outside PlanetLab was quietly dropped because the security model for a peer-to-peer CDN did not hold up. It survived anyway, because a research testbed was carrying the cost and nobody had to bill for bandwidth.

~ the unraveling ~

In August 2012 the project said active development had been stopped for a while, though the network kept running. By then the reasons to use it were being removed one at a time. Amateur video went to YouTube. Podcasts went to hosting services that handled their own bandwidth. Cloudflare's free tier arrived in 2010 and did the same job with a nameserver change instead of a URL suffix, and it did it for the site owner rather than the reader.

Sometime around April 2015 the nameservers for *.nyud.net stopped answering, and every coralized link on the internet turned into a dead one at the same moment. Nobody posted a shutdown notice. The project site sat there unchanged since 2012.

~ the next chapter ~

Asked in 2021 what had happened to it, Freedman said the platform the roughly one thousand servers ran on, PlanetLab, was end-of-lifed, and that was that. He went to Princeton, kept teaching the networking course that assigns his own paper, and later co-founded Timescale. PlanetLab itself, the thing that made a free global CDN possible for a graduate student, shut down for good in 2020.

Every free CDN tier you can sign up for today does the job a suffix used to do, with a company attached and a signup form in front of it.

@}-->--

              )
             (_)
            .-'-.
            |   |
            |   |
            |   |
            |   |
          __|   |__   .-.
       .-'  |   |  `-:   :
      :     `---'     :-'
       `-._       _.-'
           '""""""

« previous  ·  random  ·  next »