selection timeout

[Gerry.Houlder at seagate.com]

* From the T10 Reflector (t10 at t10.org), posted by:
* Gerry.Houlder at seagate.com
*

If your system is going to try the drive once, see a selection timeout, and
forever more assume the drive is not present, there is no smaller
"generally accepted" value. This is the only value guaranteed to work under
all conditions. There are a number of known conditions (e.g., reset has
just occurred) for which a device needs most of that time for recovery.

You can design a system to get by on a smaller number (like 200 usec) if
you accept some system tradeoffs.

(1) When a reset occurs, add a 1 second delay before selecting any targets.
A single initiator system should have no problem with this. This rule may
be harder to follow in a multi-initiator system where each initiator can
issue resets at any time. A better rule is to avoid resets except during
reboot (where longer response times are expected and other reboot activity
is often the determinant of reboot time anyway).

(2) If you do get a selection timeout, wait a couple seconds and try again.
This shouldn't be wasted time because the intervening seconds can be used
to initialize the targets that did answer on the first attempt. This allows
leeway for unusual target recovery actions but still reduces the scan time
immensely.

The reflector message from jmasker illustrates some of the poor system
integration decisions that can have a huge effect on system recovery.
Redundant adaptor and OS actions plus separate buses that must come up one
after the other instead of all in parallel do much more to expand reboot
time than the 250 msec selection timeout limit. If the bus scan at boot
time took only 4 seconds (which is possible with the 250 msec timeout)
instead of 30 seconds I don't think we would be hearing this complaint.
Better to take the system architects to task for expanding a 4 second
activity into a 30 second activity. This is the wrong forum for that
audience, however.



                                                                                                   
                    "Neaz, Joe"                                                                    
                    <Joe.Neaz at str        To:     "'t10 at t10.org'" <t10 at t10.org>                     
                    atus.com>            cc:                                                       
                    Sent by:             Subject:     selection timeout                            
                    owner-t10 at t10                                                                  
                    .org                                                                           
                                                                                                   
                                                                                                   
                    05/06/02                                                                       
                    02:37 PM                                                                       
                                                                                                   
                                                                                                   




* From the T10 Reflector (t10 at t10.org), posted by:
* "Neaz, Joe" <Joe.Neaz at stratus.com>
*
Is there a generally accepted value for selection timeout that is more
reasonable for current Ultra160 disk drives than the SCSI spec of 250ms? We
are trying to reduce the amount of time it takes to rescan the SCSI bus and
waiting 250ms for non-existent device ID's takes a lot of time.

Thanks,
              Joe

*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org



*
* For T10 Reflector information, send a message with
* 'info t10' (no quotes) in the message body to majordomo at t10.org