Heroku runs on top of EC2, and they get charged the same amount of money by Amazon whether your instances are idle or on fire.
Is the latter part of that true? (I'm not playing Devil's Advocate, honest!)
I'd assumed the lion share of Heroku's profits were coming from clever arbitrage of EC2 instances rather than markup alone.
For example, if they have 1000 users paying $108 per month for 4 dynos, they don't keep 4000 "dynos worth" of EC2 instances running constantly. Instead they can see the overall load and maintain that lower level of EC2 instances plus some "slack." When traffic increases among apps, the slack can be rapidly configured to pick it up. Heroku would then make not only a markup on EC2 but also profit from the difference between the maximum load and actual load. (Isn't this how cloud scaling systems work anyway, right? Heroku could/might be doing that over thousands of apps for collective benefit.)
Perhaps I'm over-thinking how I'd be trying to squeeze profits out of a Heroku-type system though ;-)
I was under the same impression. If your app has been idle for a long time it takes a few seconds for the first request to complete, which leads me to believe they shut down dynos for idle apps. But it's possible that's only for the free apps.
I know a few of the guys at Heroku and I'd be gobsmacked if they weren't doing stuff 100x smarter than I mentioned. With that said.. I hope I'm not wrong and making them feel bad in the process ;-)
Comments
Heroku runs on top of EC2, and they get charged the same amount of money by Amazon whether your instances are idle or on fire.
Is the latter part of that true? (I'm not playing Devil's Advocate, honest!)
I'd assumed the lion share of Heroku's profits were coming from clever arbitrage of EC2 instances rather than markup alone.
For example, if they have 1000 users paying $108 per month for 4 dynos, they don't keep 4000 "dynos worth" of EC2 instances running constantly. Instead they can see the overall load and maintain that lower level of EC2 instances plus some "slack." When traffic increases among apps, the slack can be rapidly configured to pick it up. Heroku would then make not only a markup on EC2 but also profit from the difference between the maximum load and actual load. (Isn't this how cloud scaling systems work anyway, right? Heroku could/might be doing that over thousands of apps for collective benefit.)
Perhaps I'm over-thinking how I'd be trying to squeeze profits out of a Heroku-type system though ;-)
I was under the same impression. If your app has been idle for a long time it takes a few seconds for the first request to complete, which leads me to believe they shut down dynos for idle apps. But it's possible that's only for the free apps.
If they're not doing it that way already then they should definitely hire you at 10% of what you would save them every year ;)
I know a few of the guys at Heroku and I'd be gobsmacked if they weren't doing stuff 100x smarter than I mentioned. With that said.. I hope I'm not wrong and making them feel bad in the process ;-)