QAS concerns

Richard Moore richard_moore at corp.adaptec.com
Mon Sep 14 09:28:06 PDT 1998


* From the T10 (formerly SCSI) Reflector (t10 at symbios.com), posted by:
* "Richard Moore" <richard_moore at corp.adaptec.com>
*
At the last minute I had to cancel my plans to attend this week's SPI-3
meeting. Nevertheless I want to address any QAS concerns.

The two open issues that I am aware of are broadcast message timing and
fairness. Larry Lamers has done some testing that supports the
feasibility of broadcast messages so I will try to address fairness.

First of all, I don't believe there is an issue with fairness on a
QAS-only bus, if all QAS devices implement the fairness algorithm.

On a mixed bus, 97-199r9 proposes a solution to the problem of QAS
devices blocking non-QAS devices from arbitration. The idea is to
require a target which is selected by QAS, and which has its fairness
register empty, to disconnect by legacy means when it is ready to
disconnect.  (I didn't require the target to do so following reselection

because this might result in too few QAS cycles to attain maximum
efficiency).

I believe it can be shown that this scheme guarantees arbitration cycles

for legacy devices. If there are several QAS-enabled targets, then the
QAS-enabled target with lowest priority will have an empty fairness
register as soon as all initiators below it have had arbitration wins.
Even
if this device is never selected again, the device with next lowest
priority
will eventually have its fairness register cleared, too. This regression
can
be continued until the highest priority QAS-enabled target is
considered,
and if it is never selected, then we must conclude that there is no
activity
by QAS devices on the bus and that legacy arbitration can occur.

On the other hand, if one of the QAS-enabled devices that has an empty
fairness register is selected, then according to the rule it must
disconnect by legacy means. This gives any non-QAS devices the
opportunity to arbitrate.

 -> If the non-QAS devices have higher priority than the QAS devices, it

would be possible for the non-QAS devices to dominate the bus during a
long series of legacy arbitrations. Therefore it is best if only non-QAS

devices with fairness are assigned higher priorities than the
highest-priority
QAS device.

 -> If the non-QAS devices have lower priority than the QAS devices, and

both sets of devices participate in the legacy arbitration, then if a
QAS device
wins it must register all participating non-QAS devices in its fairness
register,
so fairness is extended to low-priority non-QAS devices in this fashion.

At the time I wrote rev 9, I stated that non-QAS devices should not
implement
fairness (when on a mixed bus) AND should be higher in priority than QAS

devices, but I now believe this should be changed. Based on the above
considerations, I think instead we need a recommendation that buses be
configured with the following priorities (from highest to lowest):

    non-QAS fair devices
    QAS devices (fairness required)
    non-QAS non-fair devices

I hope all this makes sense. I know this message is going out late but I
appreciate
any feedback.

--
  Richard Moore
  Adaptec Irvine Technology Center


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





More information about the T10 mailing list