thewood1
Posts: 6529
Joined: 11/27/2005 Status: offline
|
I have a hard time imagining getting the overall throughput you need in even HPC cloud configurations, as compared to a 4-5 GHz Intel processor and SSD. Unless you have a direct and very highspeed pipe, the bottleneck is going to be the transport. Not to mention the amount data of needed to move the map tiles back and forth as needed. I haven't done anything with AWS's compute capabilities with something like CMO, so I could be wrong, but every benchmark my team has run has shown that elastic compute can rival local processing on large data-intensive analytic apps and workloads, but in single apps with low to medium workloads, EC is not anywhere near cost effective for long operations. And it also became VERY transport dependent with smaller workloads. If you have a edge gateway, it becomes more feasible. But at that point, just run it locally. Its mainly the overhead that limits it. Just dropped an email to my HPC analyst. He says, based on my very poor parameters I gave him for CMO operation, M5, with potentially M5n or M5zn. I don't know enough about it myself and had a hard time laying out the operating parameters for CMO. He did mention you'll probably need 3-10 Gbs bandwidth.
< Message edited by thewood1 -- 3/6/2021 1:04:01 PM >
|