RE: Target Reset handling

["Douglas Otis" <dotis@sanlight.net>]

Ralph,

1)
I agree.  This would allow a transport monitor the decision if there was a
problem requiring a communication reset.
2)
Separate then reset function further with an PDU Op-Code for requesting a
communication reset versus a management reset and leave connections on task
management resets.  If a reset messages does not clean up state with the
transport protocol functioning, a transport protocol abort would not improve
this likelihood?  There is status within the transport protocol to determine
transport state.
3)
Recommend the protocol Keep-Alive setting to ensure a dead connection is
detected in a timely manner.  If there is no connection, there is no
conflict as how to reset.
4)
Should a reset message not kill the connection as suggested, then reply with
an Asynchronous Event indicating this request to all affected initiators.
5)
Again, although a Autosense is mandated, (Makes sense.) lack of sense data
within the Response PDU not seen as an error.
6)
Do not send status with the optional PDU Op-Code 0x45 Read Data when it will
only be used with Good Status and much of the time, not used at all.  Two
places to look for status at different times makes for more problems and
overhead.
7)
Adding an Error and End of Exchange flags to the PDU 0x05 improves
reliability of the transfer state machine.  As a result, a Client Detected
Error event, not covered by the Task Management command PDU opcode 0x02, is
added.
8)
Change the RTT Ready To Transfer to RFE Request For Exchange to avoid
confusing mnemonics.


Doug

> -----Original Message-----
> From: owner-ips@ece.cmu.edu [mailto:owner-ips@ece.cmu.edu]On Behalf Of
> Ralph Weber
> Sent: Monday, September 04, 2000 4:09 AM
> To: ips@ece.cmu.edu
> Subject: Re: Target Reset handling
>
>
> Julo,
>
>
> > How about having two different function:
> >
> > target warm-reset (connections stay) and target cold-rest
> (connections get
> > also reset)?
>
> This sounds like Bus Device Reset (connections stay) and Target Reset
> (connections disappear) to me.  Or, if there's only one LUN per target,
> then it's like Logical Unit Reset and Target Reset.
>
> Thanks.
>
> Ralph...
>
>