I had a rudimentary batch job system running. Since I was charged for each power event, I wanted to minimize the power state changes and just queued up the work that needed GPU acceleration until there was enough to justify the cost of at least an entire hour and two power state events. When the cheap, small, always on instance received all the work output from the giant beast server, it would then send a power down command using the aws cli. Yeah, like I said, I was lazy so I just sent off the request and small server just went on about it’s business. Oops.
Like I told the aws billing rep, it would have bankrupted me, ending that venture. I asked that, since it is all virtualized, that while yes I had the virtual hardware provisioned for all that time, if they could get someone to look at the actual utilization they would see it was essentially none. They got back to me and in a day or so to let me know they had forgiven the entire amount.
AWS support actually listen, do real work, and very often it seems, do what is in the best interest of the customer even as it works directly against the company’s bottom line.
I’ve been able to reduce reserved instances when the market has turned unpredictable.
AWS isn’t a pushover, not even a little bit. Dealing with them externally has just been a good experience with professional understanding and a flexible business sense.