Andre Boegelsack wrote: > Hi to all, > > I have a short question regarding which filesystem I should use in Dom0/DomU. > I've built my Dom0 on basis of ZFS. > > For my first DomU I've created a ZFS pool and installed the DomU (with OSOL > inside). During the installation process you are being asked if you wanna use > UFS or ZFS - I've chosen ZFS. The installation process was incredible slow. > > Hence, in the next DomU I used UFS instead of ZFS. And the installation > process was pretty fast. > > This leads me to the coonclusion: ZFS on top of ZFS = don't; UFS on top of > ZFS = ok > > Can anybody verify that performance issue? > > Regards > André >
No experience here, but common sense tells me this might happen due to conflicting raid configurations: Be sure to have single vdevs in your DomU. Any old filesystem would normally fit this description. ZFS is special: it does it's own striping/mirroring, and makes the usual assumptions about your leaf vdevs in order to optimize performance. My recommendation follow from my 'common sense' feeling [2] that ZFS dynamic striping (default operation, but also key in raidz/raidz2/raidz3) will make assumptions about your underlying devices, that just won't hold for virtual disks, especially on LVM/RAID sets or ZFS (ZFS being similar to LVM/RAID in this department). Very simple example: if your Dom0 stripes across two physical disks (_simple_ example...) and your DomU stripes across two virtual disks. The two virtual disks will actually _both_ be striped across the two physical disks on Dom0 ZFS. So if DomU ZFS thinks it optimizes writes (by striping across it's virtual disks, assuming that these are physicial disk units), it will actually create (possibly distant) writes on the _same_ physical device, actually hampering performance and disk lifetime. The increased seek rates and seek times will harm performance. How far the disks accesses would be spread depends on the way you allocated the virtual disks. But if the disks are preallocated, it would probably be wide apart (assuming that the virtual disks get allocated 'sequentially' [1]. [1] sequentially as in contiguous per stripe volume [2] correct me if I'm wrong _______________________________________________ zfs-discuss mailing list [email protected] http://mail.opensolaris.org/mailman/listinfo/zfs-discuss
