The basic idea of a mirrorring or raid-z whole pools over ethernet for a sync backup or storage and service failover is fascinating simple from idea. I have read a lot of blogs with similar setups over years but it never comes to a common setup idea.
Like solutions with a cluster filesystem it shares the problem that performance is very related to network performance. Main advantage is simplicity. You must not add another layer or software. Only ZFS and Comstar/iSCSI is needed and this is part of the core OS,
Z-RAID SSF is mainly an idea of a ZFS solution with a Full Storage and Service failover
when a storageserver fails completely.
current storage ideas:
Idea 1: Storage + Backup: no realtime backup or failover
Idea 2: Dual head with the same mpio SAS storage, the classical HA approach
Main advantage is performance but it does not cover a jbod storage failure so it allows HA but without a realtime backup of data to another physical storage. You also need additional HA/Cluster software for a storage and service failover.
Idea 3: Dual Head, Dual Storage with sync pools over iSCSI and HA failover of pools and services. This is the Z-Raid SSF idea. You do not need special software to switch over services like NFS or SMB as they are fully embedded in ZFS on Solarish. After a pool import on the failover system, all services are available as everything that is needed to re-establsh services are ZFS properties (Solaris CIFS only) . You must additionally switch a HA ip to keep services available on the same ip. This is Z-RAID SSF= a network based Z-Pool with Storage and Service Failover.
Idea 4: Using a real cluster filesystem may add some features but adds either a layer above ZFS or must use another filesystem. It also does not cover the service failover aspect per default.
I basically want the keep a simple approach.
Managing such a Z-RAID cluster is conceptional simple but hard enough.
Overall performance, tuning, best setup options and stability are remaining questions. I hope that Solarish or napp-it is widely used enough to get answers and experiences about that and to come to something I would call a "best use/reference Z-RAID + SSF" setup procedure.
This is similar to the All-In-One concept with HBA passthrough and NFS storage for VMs. When I came up with this idea about 7 years ago, many claimed that as a bad idea. Now such a virtualized storagesystem with this concept is a very common and proven setup idea not only on Solarish.
