HP: Tips and tricks for EMC Arrays
Standard disclaimer:  use the information that follows at your own risk.  If you screw up a system, don't blame it on me...
mailto: dkoleary@olearycomputers.com

 
 
Current Path: Home
Document Type EN
Date 1999 Mar 19
Description Storage: Tips and Tricks for EMC Arrays
Document Id KNC031199002
Search String  KNC031199002

Problem Description

What are the recommendations for using EMC Disk Arrays?

Configuration Info

Operating System HP-UX
Version 10.x and 11.x
Hardware System HP9000
Series  S700 and S800

Solution

The following generic information should assist you with EMC Arrays.  Always check a proposed configuration with an EMC representative before making any changes.
 
  1. Turn OFF Bad Block Relocation in LVM for all volume groups using this command:
lvchange -N
    Note: This option is provided in an LVM patch and is recommended for all logical volumes on EMC arrays: PHCO_12666.  If the option is not used, file system corruption can result.  The adb command can be used to check bad block directory.

    Note: This patch, as with any patch, may be superseded. Please check for the latest patch at the Electronic Support Center (ESC) by using either of these web sites:

    http://us-support.external.hp.com
    http://europe-support.external.hp.com
     

  1. Each SCSI EMC array must always have an individual, dedicated interface card.  No exceptions are acceptable.
  2. Always have a LUN 0 configured.
  3. Always have sequential LUNs, for example, 0,1,2,3,4,5,6,7. Do not skip any LUN numbers.
  4. Still, only use eight LUNs per target.
  5. Set the pvchange timer to 180 seconds.

    6. Exception:  Sybase uses asyncio, therefore the pvtimeout should be the default value of 30 and the lvtimeout should be set to 180.  For Sybase systems using asyncio the following commands should be used:
     
    1. lvchange -t 180 /dev/VGXX/lvXXX of all lvols.
    2. pvchange -t 30 of all physical volumes.
    3. Change minor number flags of /dev/async from 5 to 4. This change keeps mirroring, but turns off immediate reporting:

      4. ll /dev/async
      minor # = 0x000005 (Change from 5 to 4)
  1. The array is supported on fibre channel as a direct-connected device or through a hub.  It is also supported with fast, wide SCSI.  Fibre can use volume set addressing on fibre, thus effectively increasing the number of LUNs.
  2. On HP-UX 10.20 and above, SAM will not show alternate links in disk and file systems--a desired behavior--unless the latest SAM patch is installed and the Common Serial Number Option is set in the EMC array.  This change was made to prevent alternate links from showing up as unused disks.  Customers may call complaining that they do not see all the EMC disks.
  3. EMC should provide a map indicating which are the primary and alternate paths.  There should be only two paths to any LUN, although an EMC array can see any LUN from any controller.  I have seen systems with one primary and three alternates.  This is an EMC configuration issue; however, LVM can switch to multiple alternate links.  EMC should configure the array so that there are only two paths.
  4. HP supports Business Continuity Volumes (BCVs), which are hardware mirror splits.  When these disks are split off by using EMC commands, the disks show up with the same LVM header information. These disks can be imported into the system once the vgchgid command is run on these disks.  This is usually done to backup the data offline.  After the backup is complete, the drives are exported and remerged using EMC commands.  The vgchgid command is provided in an LVM patch and is only supported on HP-UX 10.20.
  5. SCSI Queue Depth with EMC or Galaxy arrays


    6. There is an issue with SCSI queue depth that must be considered when a Symmetrix unit is used under the following circumstances:
    - MC/LockManger configurations
    - MC/ServiceGaurd configurations that are involved in concurrent read access by a system other than the primary system
    - SCSI or fibre channel EMC Arrays or fibre channel Galaxy arrays

There is a limitation on the maximum SCSI queue depth that will work on the Symmetrix Unit.  The queue can get quite large when multiple systems are accessing LUNs through a single SCSI or fibre channel adapter, also known as Open Systems Director (OSD).

A single SCSI (or fibre channel) adapter can handle a maximum of 12 concurrent I/Os per LUN when there are 128 LUNs or less.  When there are between 129 and 256 LUNs, a single SCSI (or fibre) adapter can handle as few as 4 concurrent I/Os per LUN, depending on the actual numbers of LUNs.

The HSC F/W SCSI driver uses a maximum queue depth of 2, whereas the HP-PB (28696) uses a queue depth of 8.  Care must be taken when deciding how many SCSI Adapters to order in a Symmetrix unit to ensure that the I/O load is sufficiently distributed among the SCSI adapters.

Set the SCSI queue to 4 for EMC arrays and fibre channel Galaxies. See the MAN  pages for SCSICTL for all information.  The command must be used on an active volume group.

Example:  scsictl queue_depth=4 /dev/dsk/c6d0t0


 
 
 
Document:
URL:
Last updated: