I think it must have to do something with my VM setup because when I just do this it behaves like I expect it to behave.
rsync -avP /mnt/usb/dca.mp4 /mnt/data/
sending incremental file list
51,200,000,000 100% 199.81MB/s 0:04:04 (xfr#1, to-chk=0/1)
sent 51,212,500,081 bytes received 35 bytes 209,458,078.18 bytes/sec
total size is 51,200,000,000 speedup is 1.00
Nice find! So it looks like the filesystem is performing okay. Now that the ashift is set correctly, what happens if you use a raw disk image for your VMs?
I’m far from a storage expert, but a qcow2 image is copy-on-write. Since ZFS already is a copy-on-write filesystem, I’m assuming using qcow2 on ZFS will bring with it some performance loss due to overhead? I’m not sure of this though.
have you tried a zvol instead of using a img file?
Some good references for what you are trying to do. Make sure to match the dataset and qcow2 recordsizes.
I need new specs. I completely glossed over the first link. Ouch.
@cloudstone I will give a RAW disk image a try just to see if it makes a difference for me.
@nx2l I am not using ZVOL just for one property alone. In order to take a snapshot of a 4TB ZVOL I actually need 4TB of available space on that pool. If I have just a file in that dataset I can take snapshots at any capacity.
@Log Thanks for sharing, it was an interesting read.
With ZFS there isn’t really a performance advantage to preallocating space. It’s a copy on write filesystem.
even if you dont have compression on… i dont think it does that, but i cant be 100% sure. so i wont fight you about it.
Yea that doesn’t make sense, snapshots don’t take up hardly any space at all to take. Changes after a snapshot do.