Global Identifiers for Initiators

Charles Monia monia at am.shrmsg.shr.MTS.dec.com
Tue Jan 3 12:48:22 PST 1995


Hi Ralph:


You wrote in part:

>....Initiators, on the other hand, can have one family move out and a whole
new 
>family move in more or less on a whim.  It's done by shutting down operating 
>system one and booting operating system two.  A shutdown and reboot of the 
>same operating system has many of the same properties.  So, does the global 
>identifier represent the running operating system and all its running applica-

>tions?  Or, does the global identifier represent the box that the operating 
>system is running in?

The global identifier is meant to represent the physical device -- nothing
more.

>My experience with VMS Clusters suggests that the latter type of representa- 
>tion is not very useful.  But, the former case is very difficult to define 
>and implement.  For example, the former case requires the operating system to 
>define and present its new global identifier (that is different from the last 
>boot) very early in the booting process.  Also, the operating system's global 
>identifier must be the same on all access ports, which might get very ugly 
>depending on which protocol layer defines the transmission methods for the 
>initiator's global identifier.

As I read it, Bob's proposal is a generalization of the dual-port extensions
passed for SCSI-3. As such, the capabilities provided, although modest in
comparision to VMS clustering, are considered to be useful == especially the
ability to supersede or override a device reservation when an initiator suffers
brain damage. 

It seems to me that the issues you raise are related to the problem of
coordination and cooperation among initiators. In my opinion, such issues are
outside the scope of SCSI.

Charles
 





More information about the T10 mailing list