Date:     Jan 06, 1993                      X3T9.2/93-002 rev 0


To:       X3T9.2 Committee (SCSI)

From:     George Penokie (IBM)

Subject:  Summary of SCSI Disk Array Items Submitted to the
          RAID Study Group

This is a summery of information received by the SCSI RAID
study group in response to a request for information on how
SCSI could be made more disk array friendly.

The following companies contributed the information contained
within this document:

-Ciprico
-Digital Equipment
-Formation
-IBM
-NCR

1.0 Glossary

Set - A group of devices which have a single mapping
characteristic.  Sets are independent from one another.

2.0 Host Interface Requests

A SCSI disk array should have a ability to on command:

-Rebuild a device or a range of LBAs.
-Disable a device or a range of LBAs.
-Determine the current state of a device at any time.
 Examples of some of these states are:

 =Spare drive status
 =Condition of the device
 =Condition of the components
 =Indication of any background activity

-Change the configuration of a device.
 Examples of configurable aspects are:

 =Error reporting
 =Capacity per set
 =Check data mapping
 =Data mapping
 =Set size (Number of devices and/or LBA range)
 =Number of sets
 =Device timeout values

 =Selecting if remap/reconstruction/data resync is done:
  +as a background operation,
  +only for data which is accesses affected areas, or
  +only under control of the host.

-Provide statistical information.
 Examples of statistical information are:

 =Number of device failures
 =Number of I/O operations per set
 =Number of I/O operations per device
 =Number of read/write errors
 =Number of read/write parity errors
 =Number of power cycles

-Do diagnostics for a single device, a single set, or all
sets.

A SCSI disk array should provide reporting of the following
conditions:

-Indication of which device failed
-Multiple failures on multiple devices for a single task
-Data error recovery which used check data
-Indication of a need for service
-Indication of a status change

A SCSI disk array should provide rules for addressing itself
and all its devices.

A SCSI disk array should provide for expanded defect list
reporting.

3.0 Responses to Issues

I1 - Would like to see a longer FRU code field.
A1 - The Inquiry command VPD page option allows any length or
     number of FRU codes.

I2 - Would like to see a vendor unique device matrix.
A2 - Vendor Unique functions are not defined in the standard.
     The ability to do anything is covered under vendor
     unique operations.

I3 - How should the Read/Write Long Command be handled in a
     SCSI array?
A3 - The Read/Write Long Command should be rejected when sent
     to the RAID LUN.

I4 - How should the read defect list be merged in a SCSI
     array?
A4 - This should not be a valid operation when sent to the
     RAID LUN.  However, when sent to a non-RAID LUN it
     should be handled no different than as defined in
     SCSI today.

I5 - How should the Format defect list be handled in a SCSI
     array?
A5 - This should not be a valid operation when sent to the
     RAID LUN.  However, when sent to a non-RAID LUN it
     should be handled no different than as defined in
     SCSI today.

I6 - How should different drive types be handled in a SCSI
     array?
A6 - Drive types have no meaning when sent to a RAID LUN.
     However, when sent to a non-RAID LUN it should be
     handled no different than as defined in SCSI today.

I7 - How should ECA or ACA be handled?
A7 - This is handled in the Queuing Model as much as possible,
     any deeper control could require the standard to dictate
     implementations.

The following issues are considered outside the scope of the
SCSI RAID Study Group:

-Performance Measurements
-Performance Benchmarks
-RAID Definitions/Implementation Criteria
-OS limitations including:
 =Multiple logical units per array
 =Logical unit addressing with gaps in LUN map
 =Dynamic creation and deletion of LUNs
 =Adding non-existent logical units
 =Non-standard block sizes (ie Not 512 bytes)
 =Intelligent support of command queuing