3 ms·
I tend to agree with you on AWS. I swear by it. But in this case I would probably go at it this way... How many hours can I run the machine before it costs mor
by throwaway2016a 10y ago
I tend to agree with you on AWS. I swear by it. But in this case I would probably go at it this way...
How many hours can I run the machine before it costs more than building one? Probably a month? (random guess) Will I be running it longer than that?
If yes, build the machine, if no, just rent it from AWS.
I can't even imagine a scenario where I run this 24 hours a day 365 that I wouldn't build out a Hadoop cluster or similar for.
- zymhan 10y agoWell, the advantage of AWS is that you don't have to run it 24/7 unless your workload requires it.
- throwaway2016a 10y agoExactly, so say you are running it 3 hours a day every day and it costs x an hour and you want an ROI on your hardware of 3 months. If y is the cost of building it... if 3x * 90 > y build else rent So at $13.338 (assuming no reserved instance) if y is less than $3600 you make your money back in 3 months. Of course a machine with those stats will not be $3600.
- eloff 10y agoYour math is off. At $13.338/hour you're looking at just shy of 10 grand per month ($9763 for a month of 30.5 days.) It looks like one of these bad boys would set you back about $40000. So you'd break even at 4 months. If you're going to go for the 3 year reserved instance, with $100K upfront, you'd be way better off on capex and opex just buying and colocating the thing (not considering other expenses.) The bigger you grow the more the balance tips towards do it yourself because your costs for colocation, managing hardware, etc amortize out better with scale. AWS gets a little better with scale but not nearly as quickly.
- throwaway2016a 10y agoMy math isn't off... it's for 3 hours a day and I state that in my post (perhaps not clearly enough). Your number is for 24. The point was if you're not full utilization it could make sense.