Hi Jim, way below...
Jim Dunham wrote: > Ellard, > >> To Jim and PSARC >> From Ellard Roush >> Subject iSCSI Target PGR directory for ZVOLs >> [PSARC/2009/168 Self Review] >> >> The Sun Cluster product uses SCSI-3 PGR for both Fencing >> and Quorum operations. Sun Cluster also uses SCSI-2, >> but that is not a subject of this case. >> >> At this point in time, a ZVOL is only accessible from >> one machine at a time. ZFS enhancements happen regularly. >> If a ZVOL can be accessed by multiple machines at one >> time please let me know. > > At this time a ZVOL can never be accessed from more than a single host > at a time. This is an artifact of ZFS, in that a ZFS Storage Pool can > only be imported on a single host at a time, and ZVOLs are created out > of ZFS filesystems, which are created out of the ZFS Storage Pools. > > This PSARC case and related change is based on the fact that a ZFS > storage pool can be exported from one node, imported on another, and > although the ZVOL moves between storage nodes, the ZVOLs associated > SCSI-3 PGR data (if present) was left behind. > > There will be an forthcoming iscsitgt man page describing this change, > and how to make the SCSI-3 PGR data follow the ZFS Storage Pool, and > thus any ZVOLs created out of this storage pool. > >> If a ZVOL could be accessed by multiple machines >> concurrently, the ZVOL could be used in place >> of Solaris Volume Manager. This would be valuable >> in a high availability cluster. >> >> Once a ZVOL can be accessed by multiple machines, >> we would need the ZVOL to support SCSI-3 PGR >> for Fencing and Quorum. The one-pager does not >> mention whether this project will support >> SCSI-3 PGR operations from multiple nodes >> concurrently. Please explicitly identify >> whether this will be supported. >> >> Naturally, we would like to see SCSI-3 PGR >> operations on a ZVOL supported for >> multiple nodes concurrently. > > SCSI-3 PGR command are supported with iSCSI Target Daemon LUNs, and > soon all SCSI Targets in COMSTAR. Of course COMSTAR is storage node, > and there is no support for a COMSTAR storage node to also be a Sun > Cluster node. I'm a bit confused by this statement. Is there some issue you see here that would preclude a cluster node from being a COMSTAR storage node? This is, in fact, the config that we use for Solaris iscsi Target currently in our Colorado 2 node shared-nothing config. Thanks, Dan > > Jim Dunham > Engineering Manager > Sun Microsystems, Inc.
