Sounds like a bug we hit at AWS a while back. The clocks on several of our instances started running at incorrect speeds. One was gaining something like 10 seconds every minute or something ridiculous like that. The rate was steady, just very far off from one minute per minute.
The Linux timekeeping system can normally deal with a clock that is steady but not the right rate by learning what correction to apply, but these were far enough off to be well beyond the maximum correction.
This only affected particular instance types, so changing to a non-affect type could fix it, but all the working types were more expensive than the types we were using which was annoying. Sometimes you could change to a working type then back to the buggy type and end up with one that did not have the problem.
There were a few people asking about this in the support forums, but apparently it was only hitting cheapskates like us who had not paid extra for support. Finally, someone paid for incident support and opened a ticket and Amazon quickly found and fixed the problem.
The Linux timekeeping system can normally deal with a clock that is steady but not the right rate by learning what correction to apply, but these were far enough off to be well beyond the maximum correction.
This only affected particular instance types, so changing to a non-affect type could fix it, but all the working types were more expensive than the types we were using which was annoying. Sometimes you could change to a working type then back to the buggy type and end up with one that did not have the problem.
There were a few people asking about this in the support forums, but apparently it was only hitting cheapskates like us who had not paid extra for support. Finally, someone paid for incident support and opened a ticket and Amazon quickly found and fixed the problem.