Comments on SDA States

Doug, dtn 237-2145 Flames to NL: 07-Jan-1994 1100 hagerman at starch.enet.dec.com
Fri Jan 7 08:00:29 PST 1994



Date:           January 5, 1994                 X3T10/94-___ Rev 0
To:             X3T10 Committee (SCSI)
From:           Doug Hagerman (Digital)
Subject:        Comments on SDA States (93-191)

This proposal is an expansion of the SDA States described by George
Penokie. Given the complex nature of RAID devices or any general
SCSI subsystem, it seems to me that the initiator used to configure
and control the subsystem must have considerable visibility into the
current state of the subsystem. The following example traces one
possible sequence of events in the use of a RAID subsystem, and shows
why the number of states needs to be expanded over George's proposal.

Event                           New State
-----                           ---------
Power on to subsystem           Self-test in progress
Self-test completes             Ready to handle INQUIRY command
Self-configuration completes    Ready to handle any command,
                                P-addressing valid (only)
Define first extent             At least one extent available
Define first redundancy group   At least one redundancy group available,
                                R-addressing valid
Define first volume set         At least one volume set available,
                                V-addressing valid
Enable RAID operation           Normal quiescent operating state

Disk failure                    Degraded redundancy
Initiate rebuilding on spare    Degraded redundancy, rebuild in progress
Rebuild complete                Normal redundancy, degraded performance
                                (because of use of spare)
Failed disk replaced            Normal redundancy, copyback in progress
Copyback complete               Normal redundancy, normal performance,
                                erasing spare
Spare reconfiguration complete  Nomal quiescent operating state


Many of the above states are volume set states, but P-extent,
P-LUI, redundancy group, spare, and overall SDA states are intermixed.
Considering that a RAID subsystem of moderate size might have several
volume sets, dozens of redundancy groups, and hundreds of P-LUIs,
it is clear that the state description needs to be more comprehensive.
Perhaps we need a separate table for each of these, as proposed below.

x.0 SCSI-3 Disk Array States

The following sections describe the possible states for each
of the addressable components of a RAID subsystem.

x.1 DACL States

An SDA consists of one or more addressable DACL devices at one
or more SCSI device addresses. Each DACL can report the following states.

State                           Condition
-----                           ---------
Self-test in progress           Power is applied to system, but DACL is not
                                yet ready to respond to any command.
Ready to handle INQUIRY command Self-test has completed, DACL is ready
                                to respond to INQUIRY command.
Ready to handle any command     Any self-configuration operations
                                (possibly requiring media access) are
                                complete, DACL is ready to respond to
                                any command.
Hung                            DACL is not responding to commands.

x.2 Volume Set States

A DACL implements one or more volume sets at one or more LUN addresses.
Each volume set can report the following states.

State                           Condition
-----                           ---------
Not available                   The volume set has been defined, but
                                is disabled.
Active                          The volume set is available for normal use.
Degraded performance            Within the volume set some component
                                is in a state that causes degraded
                                performance. All data is still fully
                                protected.
Exposed                         Within the volume set some component
                                is in a state that causes the loss of
                                redundancy. All data is still valid.
Failed                          Within the volume set some component
                                is in a state that prevents a guarantee
                                that data will not be lost.

x.3 Redundancy Group States

A DACL implements one or more redundancy groups at one or more LUN addresses.
Each redundancy group can report the following states.

State                           Condition
-----                           ---------
Not available                   The redundancy group has been defined, but
                                is disabled.
Active                          The redundancy group is available for
                                normal use.
Degraded redundancy             Something has failed, causing the protection
                                of the data to be lost, and the system
                                does not have a resource needed to provide
                                protection. No data has been lost.
Rebuild in progress             Protection is being rebuilt onto a spare
                                or repaired disk. Protection has been lost,
                                but the system is working to rebuild it.
Degraded performance            A component that affects performance has
                                failed, or a lower-performance spare
                                disk unit is in use. In this state all
                                data is still fully protected.
Copyback in progress            Data is being copied back to a repaired
                                disk. All data fully protected. The system
                                is working to provide full performance.


x.4 P-LUI States

A P-LUI is an underlying device controlled by the DACL. P-LUIs include
the disks used to store the data, plus any other devices used in the SDA
that can be addressed with the physical addressing mechanism.

Since each P-LUI is an addressable physical device in the system,
its state is simply the state of the physical device. SCSI devices do
not report "state" as such, but the state can be inferred from the
data returned in response to commands such as INQUIRY, TEST UNIT READY,
and REQUEST SENSE.

[Not sure about this. Is the DACL capable of reporting how it is
using each P-LUI, e.g. "P-LUI is in use in R-LUI n"?]

Each redundancy group can report the following states.

State                           Condition
-----                           ---------
Not available                   The redundancy group has been defined, but
                                is disabled.
Active                          The redundancy group is available for
                                normal use.

x.5 Spare States

A spare is an addressable [something] that can be used as a spare.

Each spare can report the following states. Note that there is no
provision for reporting the history of prior use of a spare. It is
assumed that a spare is available for temporary use but that when the
original device is replaced, the data will be copied back to the
original device and the spare will be erased and made available again.

State                           Condition
-----                           ---------
Not available                   The spare has been defined, but is disabled.
Available                       The spare is available for use.
In use                          The spare has been allocated for use
                                as a spare.





More information about the T10 mailing list