Measuring Feedburner's latency

Thursday, January 08, 2009 by Dave Winer.

Yesterday I listened to a Gillmor Gang podcast that focused on one issue -- how much time does it take Feedburner to reflect the changes in a feed they're hosting. Steve had some evidence that it was taking as much as three hours for it to reflect changes in his feed at techcrunchit.com.  

Being an engineer, I wondered what was going on, so I constructed a test with a feed to see what Feedburner would do with it. 

Here's the original feed and here's the Feedburner version. 

Here's what my test does. Every minute it reads the Feedburner version and compares it against the original. If they don't match, it does nothing. When they do match, it notes the time in a log, generates a new version of the test feed and repeats the process.  

I'm going to let the test run for a few hours and then make one change -- I'll ping their server when I create the new version. 

And of course I'll report the results here when they are available. 

A note: I ran the test overnight and got what to me are astonishing results. Feedburner never noticed the change in the original feed. Anyone who was subscribed to it would not have known there had been news. I couldn't believe this, I felt there had to be a bug somewhere in my test, and it could be that there is. That's why I'm re-running it this morning while I'm working and can keep an eye on it. 

Update at 11:10AM Pacific: First results after running the experiment for almost 2 hours: It took the following amount of time for Feedburner to reflect a change in the original feed: 24 minutes, 31 minutes and it's still returning old results after 61 minutes. This is without pinging. 

Update at 2:20PM Pacific: Here's a table that summarizes the results.