How much RAM do you think they have? I think a top-notch scientific machine from literally 20 years ago might have like 4GB RAM, maybe a bit more. I’m not sure about non-control-plane kubelet runtime overhead, I see online sources showing numbers as little as 30MB, and as big as 1GB. But even with 1GB you will still have 3GB for work.
For regular k8s a few years ago (in my own tests) it hovered around 500-600 megabytes, and some cpu as well (can’t say how much would it use on machines this old) so the thing that does nothing but controls what will run on a machine consumes like 1/8th of ram. Also their infra sounds as static as it gets and machines themselves are not pets - they are family members at this point, what would the benefits of k8s be?
Compared to what? Manually running software on machines? Automatic distribution of work, automatic networking, linking/transfer of resources, etc. I think their scenarios are often oneshot tasks that need to run somewhere and produce results that get used in other tasks or aggregated somewhere, etc.
It actually requires not an insignificant amount of resources for 20 y.o. machines?
How much RAM do you think they have? I think a top-notch scientific machine from literally 20 years ago might have like 4GB RAM, maybe a bit more. I’m not sure about non-control-plane kubelet runtime overhead, I see online sources showing numbers as little as 30MB, and as big as 1GB. But even with 1GB you will still have 3GB for work.
For regular k8s a few years ago (in my own tests) it hovered around 500-600 megabytes, and some cpu as well (can’t say how much would it use on machines this old) so the thing that does nothing but controls what will run on a machine consumes like 1/8th of ram. Also their infra sounds as static as it gets and machines themselves are not pets - they are family members at this point, what would the benefits of k8s be?
Compared to what? Manually running software on machines? Automatic distribution of work, automatic networking, linking/transfer of resources, etc. I think their scenarios are often oneshot tasks that need to run somewhere and produce results that get used in other tasks or aggregated somewhere, etc.
There’s decades of effort already done on this kind of software in science. Idk why Kubernetes would suddenly be needed for that.