HP OpenVMS Systems

ask the wizard
Content starts here

Cluster, Multihost SCSI, RAID? (KZPAC)

» close window

The Question is:

We have an Alphaserver 2100 cluster that uses DSSI disks in storageworks
boxes and HD05 controllers.  This still works but is very old and we need a
significant increase in disk capacity.  The old system uses OpenVMS
shadowing for redundancy.
We bought a BA356 storageworks box full of large ultra-SCSI disks and two
KZPAC-AA raid controllers with appropriate cables and personality module.
The raid system works perfectly, but cannot be used in a cluster, that is,
disks connected to both Alphase
rvers at the same time.
As a fallback, can we use KZPBA-CA non-raid controllers and OpenVMS raid
software as a SCSI cluster disk system?
Thank you
The OpenVMS version is V7.1-2, but we are willing to upgrade anything as

The Answer is :

  The OpenVMS Wizard will here assume that the disk hardware redundency
  provided by volume shadowing (also refered to as disk mirroring) is
  the core reason for the particular application of RAID, and not disk
  striping nor larger volume capacity requirements that can also be
  achieved with various RAID configurations.
  As a general rule, the SCSI buses emanating from backplane-based RAID
  controllers (eg: KZPAC RAIDarray 230) are not supported for multihost
  SCSI operations on OpenVMS.  (Host-based volume shadowing can be used
  across these backplane-based RAID controllers, but that may or may not
  be a desirable storage configuration in your particular environment.)
  As a general rule, host-based SCSI controllers that support SCSI TCQ
  (Tagged Command Queuing) generally are supported for multihost SCSI
  configurations.  Additional configuration requirements:
    o one or more cluster communications interconnects (eg: FDDI,
      Gigabit Ethernet, CI, etc) are a requirement for any multihost
      SCSI configuration, as the OpenVMS Cluster System Communications
      Services (SCS) cannot communicate host-to-host over SCSI.
    o SCSI widgets that are not transparently tolerant (re-entrant)
      of SCSI bus resets (eg: magnetic tapes) are not permitted on
      a multihost SCSI bus.
  As a general rule, the SCSI buses emanating from various HSZ, HSJ,
  and similar storage controllers can themselves be configured into
  specific multihost environments -- for instance, a SCSI bus emanating
  from an HSZ series controller can be connected to that of another HSZ,
  with appropriate firmware revisions loaded in the controllers.
  In addition to disks, HSZ-series and similar (non-backplane) disk
  storage controllers are usually supported for use on multihost SCSI
  buses, and -- with the appropriate OpenVMS releases, controller or
  drive, and controller or drive firmware.  Various disk storage
  controllers can also span across host-based multihost SCSI buses,
  permitting failovers to transparently occur across SCSI buses.
  The KZPBA-CA series controller (PCI to wide UltraSCSI) can be used in
  multihost SCSI OpenVMS Cluster configurations, though (as mentioned
  earlier) one (or more) additional SCS communications interconnects
  are also required.  While OpenVMS Alpha V7.1-1H1 was the first release
  with UltraSCSI support -- an ECO kit is available to retrofit UltraSCSI
  controller support for non-multihost configurations to OpenVMS Alpha
  V6.2-1H3 -- OpenVMS Alpha V7.1-2 and V7.2-1 (and later) are the OpenVMS
  releases recommended as a minimum by the OpenVMS Wizard.
  Also be cogniscent of and carefully follow the requirements in the
  OpenVMS documentation and in the OpenVMS FAQ around the settings for
  the VOTES and EXPECTED_VOTES parameters.  Incorrect settings for these
  system parameters can lead to cases of severe disk data corruptions.
  OpenVMS Alpha V7.1 and later -- and specifically the addition of
  support for SCS port allocation classes in these releases -- makes
  configuring and operating a multihost SCSI environment far easier.

answer written or last revised on ( 6-JAN-2000 )

» close window