razique.mahroua at gmail
Aug 17, 2012, 7:21 AM
Post #1 of 1
working on the new implementation of nova-volume inside my cloud, I'm trying to avoid the implementation I had during a year, the "classical" approach, consisting in creating a VG on a partition, that last being on a SAN.
Here are few new implementations I was thinking about
• 1 - Install on every compute-node nova-volume, meaning reducing risks to break the instance which uses a volume in case the network cuts down (power outage, network failure, bug) - that wouldn't guarantee full HA but would reduce the risks proportionally to the number of nodes.
In such setup is it possible to specify for a node to use it's own volumes for the instances running on it ?
• 2 - Use a shared storage (MooseFS in my case) for storing the VG as a file, meaning only one big file shared across the compute-nodes. Does that sound realistic ?
• 3 Install CLVM and using maybe CARP for a failover between serveral VG (as files) - but keep the exposition from the controller -that wouldn't prevent though instance to have erratic behaviors in case of network loss
What is the implementation you consider the safest so far ?
Nuage & Co - Razique Mahroua
razique.mahroua [at] gmail