Louis Guillaume <louis%zabrico.com@localhost> writes: > I'm setting up my dom0 and would like to organize all the disk space > for the hosted domains in one place (/srv/xen). And I want to do it in > the most flexible and/or efficient way. To me, that means files for the virtual disks, accessed via vnd. > o What benefits do we gain from using vnd devices in our xen > configurations over regular files? It works. Basically, on a NetBSD dom0, to provide a disk to a domU requires a block device on the dom0. This can be done with an actual block device (partition of a disk), or a file-backed vnd. > o What's the correct practice for managing vnd devices? Other than > having a script to vnconfig them each time we boot, how should > these be tracked? In your xen config, disk = [ 'file:/n0/xen/foo-wd0,0x1,w'] makes that file be the contents of xbd0. > Any help clarifying this would be great. I'm using netbsd-5/amd64 for > dom0 with a recent xenkernel33 and xentools33. When you start up a domU, the scripts create a vnd and attach it to the file. The advantage of the file/vnd appraoch is that you can resize disks with dump/restore on the dom0 and new files, and they live in a big fs. With partitions it is messier. But with partitions you skip dom0 fs overhead, which is perhaps 10%. When you create the first disk, so so by dd'ing zeros onto the entire planned size. vnd/xen does not play well with sparse files, and you want it all pre-allocated for good locality anyway. I am unaware of how to make xen virtual disks work with a file without vnd. If you are hvaing trouble, please post your config file.
Attachment:
pgp2T7ncXoBUs.pgp
Description: PGP signature