SCSI/ATA Power Management

Introduction

Power management for SCSI and ATA devices becomes important in mobile and large server environments. Most of the devices discussed here have interfaces based on (draft) standards at www.t10.org (SCSI) and www.t13.org (ATA). This page attempts to give an overview of the terminology and the mechanisms available.

Details

The SCSI Primary Command draft (SPC-3) defines commands common to all SCSI devices. The SCSI Block Command draft (SBC-2) defines commands  used primarily by disks. The MultiMedia Command draft (MMC-4) covers optical storage devices such as CD and DVD drives. The Serial Attached SCSI draft (SAS-1.1) covers a transport protocol that can convey both SCSI command sets (via SSP) and the ATA command set (via STP). In some cases SCSI command sets are carried on ATA transports. For example most CD/DVD drives are physically connected to an ATA bus (transport) which uses its "Packet Interface" (hence ATAPI) to convey the SCSI MMC set.

The power management state that a device is in depends on several factors:
Not all of the draft standards have the same set of power management states (see table below). At least the naming of states seems to be consistent.

SPC-3

 		SBC-2
SCSI disks
 		MMC
CD/DVDs
 		SAS (SSP)
 (s)ATA
disks
 		Notes
Powered_on
 Powered_on


"This state consumes zero time."
Active
 Active
 Active
 Active
 Active
 Medium accessible with least delay. Maximum (average) power requirement
Idle
 Idle
 Idle
 Idle
 Idle
 Compromise between Active and Standby
Standby
 Standby
 Standby**
 Standby
 Standby**
 Medium typically stopped (spun down). No initializing command required before next media access (but it may need up to 30 seconds to respond)

Stopped**
Stopped
Similar to Standby but an initializing command [START STOP UNIT (start=1)] is required


Sleep
Sleep
 Device reset or power cycle needed before device will accept further commands



Active_Wait
 ***
 A SAS link layer NOTIFY(ENABLE SPINUP) is required (in SSP)



Idle_Wait
A SAS link layer NOTIFY(ENABLE SPINUP) is required (in SSP)

Those states marked with "**" are the resultant power condition state after a START STOP UNIT (start=0) SCSI command has been executed. For (s)ATA devices the SCSI-ATA Translation draft (SAT at www.t10.org) specifies a translation to the STANDBY ATA command in this situation.

The state marked with "***" is for SATA disks connected via the SAS transport's STP. The  bridge STP/SATA phy in an expander may be held in SpinupHold state after power up. This will be reported in the "negotiated physical link rate" field of a SMP DISCOVER response. It is then up to a SMP application client to issue an SMP "phy control" function to perform a link reset or a hard reset on the STP/SATA phy which will result in the SATA disk spinning up. How this could be done from SAT is currently unclear.

For SAS disks (connected via SSP) in Active_Wait or Idle_Wait state a media access command will result in a NOT READY sense key with an additional sense code of LOGICAL UNIT NOT READY, NOTIFY (ENABLE SPINUP) REQUIRED. If the SCSI subsystem in the host computer receives that sense data (or the equivalent message from a SATA bridge via STP) then it needs to either:

  1. use some algorithm within the kernel to determine when to instruct the SAS link layer in the HBA to issue one or more NOTIFY (ENABLE SPINUP) primitives
  2. alert the user space, potentially via the hotplug mechanism
  3. in the case of a pass-through interface, pass back the sense data to the user space application
Options 2 and 3 are only practical for control if some mechanism exists for a user space application to instruct the SAS link layer in a SAS HBA to issue one or more NOTIFY (ENABLE SPINUP) primitives.

There is not much information about what power state real world SAS disks power up in. The Hitachi Ultrastar 15K147 SAS disk product manual has a "RNOT" bit in the unit attention mode page (which is vendor specific) for whether a NOTIFY (ENABLE SPINUP) is required or not. With RNOT set (the default) the disk powers up into ACTIVE_WAIT state (i.e. a NOTIFY (ENABLE SPINUP) is required), and if RNOT is clear then the disk powers up into ACTIVE state. Some notes about the power management of this disk: it does not implement the Idle condition timer and while it does implement the standby condition timer, the minimum standby condition timer period is 60 minutes, irrespective of the value given to the Power Condition mode page (via a MODE SELECT command). Perhaps Hitachi are concerned about too many spin-down/spin-up cycles over the service life of the disk.

Enclosures holding SAS and SATA disks will most likely include SAS expanders which are also capable of issuing the SAS link layer NOTIFY (ENABLE SPINUP) primitives. Hence logic in an enclosure could control the staggered powering up of disks. That could lead to some competition between multiple initiators (in host computers) and enclosures for spinning up disks.


Doug Gilbert (dgilbert@interlog.com)
Last updated: 21st November 2005