tronic wrote:
Linode service is pretty good on the whole. There's the occasional UML bug or rare network burp, sure. So at best, Linode.com's uptime is actually somewhere in the ballpark of 99.990-99.993% on an individual host level.
I don't get the impression that this service was designed for a 24x7 100% availability setup; it was more targeted towards the 99.xx% crowd.
I agree with much of your post, viz. the suggestion that people should not be looking at Linode as a 100% uptime, high-availability solution, but I disagree with your figures here. In my experience, when all factors are taken into consideration, Linode is more like a 97% uptime system. The load spikes which make your Linode unavailable, even though the host system is still running and connected, combined with the kinds of unexpected downtimes such as are being discussed in this thread, which although not common are not unheard of either, mean that Linode is not a 99.9% uptime host either.
The load spike problem has been improved considerably by caker with the recent kernel patches. But if you look at my load stats on
http://www.ischo.com/munin/ischo.com/eva.ischo.com.html you'll see that even now there are a couple of times per week when a significant spike (anything over 2 has an effect, anything over 3 makes the Linode almost inaccessible) interrupts service for a few minutes. In my 1.5 years with Linode I've seen the problem get much less significant but on average, I can give a rough estimate of 97% uptime given this issue.
Quote:
Could Linode.com clarify the 100% figure better on the web site? Certainly. Could add some wording indicating that this was between TP and Linode.com, and not between Linode.com and the individual Linode.com customers. Better yet, add some explicit wording mentioning actual availability between Linode.com and the customers were on a best-effort basis.
This would be very helpful. I hope that Linode.com can respond to this request.