I ran a couple data centers on either coast each burning about half a megawatt, which isn't that much in the grand scheme of thing. My employer at the time offered a SaaS product to Fortune 100 companies. The product required a realtime kernel and QOS in the network, neither of which is supported in AWS.
I did a fair bit of math comparing costs. Typically, you could pay off the CapEx with 6 months of usage in AWS.
IIRC we had around $8m in salaries for the ops org. I'm not sure we would've needed a great deal fewer people having all that infra in AWS.
The real problem comes from trying to find a team of people willing to engineer your service for failure. That kind of attitude just doesn't exist in the world of IT, so anyone would be hard pressed to put a team together.
Good luck finding a network engineering team that knows BGP well enough for such a thing. Good luck finding a team of engineers that know Openstack well enough to keep it running. Good luck finding folks to run Ceph. Good luck getting all that right.
If you could, though, there are certainly savings to be had.
We aren't talking about a datacenter to support one product. We are talking about an entire platform for building and operating any number of products on top. It isn't comparable.
Nah.
I ran a couple data centers on either coast each burning about half a megawatt, which isn't that much in the grand scheme of thing. My employer at the time offered a SaaS product to Fortune 100 companies. The product required a realtime kernel and QOS in the network, neither of which is supported in AWS.
I did a fair bit of math comparing costs. Typically, you could pay off the CapEx with 6 months of usage in AWS.
IIRC we had around $8m in salaries for the ops org. I'm not sure we would've needed a great deal fewer people having all that infra in AWS.
The real problem comes from trying to find a team of people willing to engineer your service for failure. That kind of attitude just doesn't exist in the world of IT, so anyone would be hard pressed to put a team together.
Good luck finding a network engineering team that knows BGP well enough for such a thing. Good luck finding a team of engineers that know Openstack well enough to keep it running. Good luck finding folks to run Ceph. Good luck getting all that right.
If you could, though, there are certainly savings to be had.