[one-users] Best Practice - Storage Backend

[InterNetX - Juergen Gotteswinter]

 Hello Nebula Users,

i am curious what others are using as Storage Backend, since i spent
plenty of time trying to get block LVM or FS LVM to work reliable. CLVM
seems like a cure to me, and from what i could find through Google so
far i am not the only one who experienced things like (everything on a 4
Node Cluster, seperate Management System for Opennebula/Sunstone -
Nebula 4.6 and 4.7.80):

- whole Cluster dying / hangs (no more VG/LV Operations possible) if one
Node decides to die
- unreliable if several VM get deployed at once (didnt investigate further)

i tried:

- CentOS 6 based KVM Hosts, cman/clvmd based Setup
- CentOS 7 based KVM Hosts, Corosync/clvmd (feels little bit better, but
still not like something i whould consider for Production)


What i wanted to avoid is Using NFS as Backend, the Experience over the
last few years with NFS as Backend for Images is not the best in term of
Performance (if you want to make sure your data stays consistent, eg
using sync NFS).

Block Device, in form of a LV or similar things like ZVOL is where we
got best experience over years with other Virtualization Platforms.
Thats something i want to keep for sure.

Since the Hostsystems are pretty huge, there wont be just run a hand
full of VM on this Setup (1TB Ram avail. in this Setup for the
beginning). So one has to keep in mind that its not a small scale Setup.

Every shared experience / or maybe hint how to deal right with CLVM is
highly appreciated.

Cheers,

Juergen