Re: iSCSI: Canonical Targets

[Stephen Bailey <steph@cs.uchicago.edu>]

Rob,

> Surely you are aware that most medium-to-large-scale
> storage arrays present multiple targets today, usually
> in the form of multiple FC interfaces, each acting
> as a target (i.e. FC WWPN)?

Yes.  However, this behavior is not really doing what you claim.  WWPN
is a way to uniquely identify an INTERFACE.  From the standpoint of
SAM's notion of target, you are correct that an FCP interface presents
a unique `target'.

In reality (one step above SAM), these boxes present a single target
visible through multiple interfaces.  There is no independent target
addressing coordinate in FCP as there is in iSCSI or SST.  What really
represents the target in FCP, from an addressible entity standpoint,
is the FC Node Name.  The targets presented on different ports by the
same box have different N_Port_Names, but they all have the same
Node_Name.  That means if you change the contents of a disk LUN
through one port, you will see the changes (eventually) of a
corresponding LUN on another port.

Furthermore, there seems a strong move afoot to modify SAM to embrace
this notion of multiple views of a target.  Seeing a single target
through multiple interfaces behavior has been the norm for enterprise
storage since ||SCSI.

A wedge driver (your host or HBA-level software) uses Node_Name, or
more likely something from the VPD of inquiry (serial number, device
Id, which probably is the FC Node_Name) to keep /dev entries straight,
not N_Port_Name.  The VPD information is available across transports,
so you can still find the target on a box with heterogeneous links.

Steph