On Limitations and Maximums

[tlaramie]

I've done so looking around but I haven't been able to find any detail on maximums that PDM can handle such as:
- Clusters
- Nodes
- VMs
- etc
I'm interested in not just the limitations but how this affects resource scaling for the system (e.g 5000 VMs would require 128GB of RAM and 1TB of disk with a max latency of 10ms). Has anyone stressed PDM enough to have an idea of where things stop to hockey stick?

 

[YennyPudge]

i dont know what issues could be, looks like memory is enough

 

[dcsapak]

there are no artifical limits in pdm itself, so the only limit is what the system can give.

mostly this is
* network latency/throughput (e.g. for remotes with many vms updating the metrics, etc. can be quite large)
* memory: from what i can remember a single resource with metrics (e.g. a vm/node/container/etc.) uses about ~1MiB memory, so 5000 vms would require about 5GiB of memory, but i'm actually not sure if this still is the case, so don't quote me on that
* for disk, i guess about the same is true, probably a bit more since there is additional data on disk vs memory (task logs, etc.) so i'd use some safe factor for planning

take this info with a grain of salt though, since the last time i actually measured it was quite some time ago

in any case, we tried to make the memory/cpu/network footprint as small as possible, so unless the environment is huge, a reasonably sized machine should get the job done

EDIT: just wanted to mention: even if these figures are correct, these are subject to change, e.g. if we need to keep more data in memory per resource for important functionality, we will do so (but of course trying to avoid using too much memory at once)