[jedsmith]

I have a bit of experience with Xen. If you're actually seeing a whole lot of steal (how much?), that's a bad sign because it means you're on a box with a lot of contention. In an ideal world, Xen should steal very little from you. I'm burning all four cores available to me on one of my personal Linodes, and the platform is barely stealing anything. Here's vmstat -s and uptime from that Linode for comparison:

       409198 non-nice user cpu ticks
     60878563 nice user cpu ticks
       166987 system cpu ticks
    811571786 idle cpu ticks
      4486779 IO-wait cpu ticks
           25 IRQ cpu ticks
        15388 softirq cpu ticks
       766577 stolen cpu ticks

    12:06:10 up 13 days, 14:11,  3 users,  load average: 4.00, 4.01, 4.05

I've had the pedal to the floor for a couple of days on the CPU, and only 766 kticks have been stolen (total) since I booted. If you're seeing a lot more steal than that, your host is working pretty hard to schedule the domUs fairly.

Wouldn't dare to assume that I know better how to run operation than you do, just sharing my experiences with Xen. Netflix had a solution to this -- unfortunate that it was necessary, but a solution nonetheless -- which was to monitor steal closely and spin up a new instance if it skyrocketed: http://blog.sciencelogic.com/netflix-steals-time-in-the-clou...

Given the opportunity, I'd like to point out that I meant no disrespect in my original comment, if it wasn't clear. I was speaking more from a generality and not about CKAN specifically, a fact lost on those mindlessly downvoting me.

[kindly]

No disrespect was taken. The hacker news coverage came as a big surprise. We like to turn any caching mostly off and we know this is a risk. This is because we do not want the possibility of any stale data as this annoying to the type of users we have. We are working on a better cache invalidation scheme but this has not been a big priority.

Your feedback is appreciated, thank you.

Edit: Our amount of steal was much much higher than that.