jedfrechette wrote: ↑Fri Aug 06, 2021 7:34 pmYep same here, we're tiny company as well (less than 10 people).
Spinning rust on the servers (TrueNAS) here. We're using mirrored VDEVS and can easily saturate the 10G network links to our workstations so I don't think going to flash would provide much benefit for us at this stage.
For some things (mostly photogrammetry) we do work directly off the NAS. The problem we run in to with other process is that some of the applications we use like to store their project data in millions of little tiny files. The network overhead associated with reading and writing all of these little tiny files is obscene and makes working directly over the network impractical. For those workflows it is better to transfer everything over locally, do the work off an NVME drive, and sync back changes.
Truenas here as well. Recently we changed to 3vdev radz2 from 6vdev mirror. To save some storage space. currently only 1G connection but it is planned to upgrade. Tiny files will always be the bottleneck. But i thought we can improve the small file access by going to flash based (or just using metadata vdevs which can store small files, it is an interesting approach to hybrid storage from the zfs community), I didn't have time to test, if it is network or storage server bottleneck. We are using register 360 and it is not that bad through network. import same, export almost same (cleaning can be slower if you have huge corecount). bundle optimization is slightly slower (automatic optimization is off always, no matter what it would just waste of time)
It is offtopic here, But it would be interesting to how other sysadmins/tech savvy peoples handle the huge amount of data when dealing with point clouds.