tape_serial

[Dal Allan]

* From the SCSI Reflector, posted by:
* "Dal Allan" <dal_allan at mcimail.com>
*
Howdy all,

I've been following the tape thread which is titled Tape on PLDA but seems 
to be on scsi at symbios most of the time though it has also turned up on 
disk_attach at dt.wdc.com. Rather than subject both reflectors to this email, 
this is going only to scsi. BTW, although it is entitled as a PLDA issue 
the problems and solutions seem more generic than FC-specific.

First up, am I listening to the wrong people? The tape vendors which have 
chatted to me seem to have a lot fewer problems with the shift to serial 
than most of the ones on the reflector(s) today. 

I've been reading the notes on tape with interest, and so far find it 
interesting that tape vendors are not firing missives. Actually, the traffic 
looks to have too many assumptions in it about the way tapes can or will 
physically behave. 

Doug's paradigm provides a model of expectations that the host has of the 
tape drive. With that information, the designer can go off and build a 
device that meets them. For years now, tape vendors have been masking 
physical problems in media movement from the host. 

I don't buy the assumption that a really high speed drive will drop itself 
back from streaming mode because of the 'fill buffer first' paradigm. The 
faster the device the more pressure there is to keep streaming because you 
are talking huge database dumps and the like. Tape guys are just as 
innovative as disk guys and host guys, they'll figure out how to meet the 
paradigm AND keep streaming. It may not be first generation, they'll be 
focusing on simple transition, but count on it to come soon thereafter. 

Don't take this wrong and assume I don't support the ideas coming out. One 
of the biggest problems that tape drives have had is that host drivers have 
been inconsistent in what they expect of the drive. What Doug has started on 
describes a model which all hosts will adopt if it gets documented in the 
profile. Wow, that's a huge step forward, the device guy can count on 
consistent behavior and if it doesn't occur can point to the profile as the 
correct way to manage the device. 

Concentrating on what the host wants to see as behavior and documenting what 
is expected of the drive looks like sufficient guidance to me. 

Trust your local tape vendor. 

 - What happens if a vendor calculates that a 500 KB buffer is sufficient to 
   sustain streaming and tells the host it can handle 1 MB? 

   In this competitive world, shaving cost is what its all about. 

 - Are you going to tell me it is non-compliant? 

   If a failure occurs and the tape drive is repositioned to the point where 
   it was before that command began execution the host can't tell that the 
   tape drive lied because the recovery will work correctly. 

As long as the focus is on the paradigms of host expectations, tape vendors 
can tell us if there are any invalid assumptions or are non-workable. 


Dal