Jump to content United States-English
HP.com Home Products and Services Support and Drivers Solutions How to Buy
» Contact HP
More options
HP.com home
Upgrading from HP-UX 9.x to 10.x: HP 9000 Computers > Chapter 4 Pre-Upgrade Tasks for All 9.x Systems

Modifying Your System

» 

Technical documentation

Complete book in PDF
» Feedback
Content starts here

 » Table of Contents

After running snoop, do the following:

  • Read the logfile /var/adm/sw/snoop.log paying particular attention to PROBLEMS (which you must be correct) and CAUTIONS (which you should be acted on or corrected). Also check /var/adm/sw/swagent.log.

  • Perform the steps that are recommended in the logfiles.

  • Read “Preparing for I/O System Changes ” later in this chapter, and the sections that follow, and perform the tasks that apply to your particular system.

  • Run snoop again until all PROBLEMS and as many CAUTIONS as possible are corrected.

    This may mean running snoop a number of times. You don't need to run snoop interactively each time. To run snoop unattended, enter a command such as:

    /usr/sbin/snoop -a -o -s petunia:/mydepot

    (This example uses a network server named petunia which has a 10.01 software depot under the directory /mydepot.)

    NOTE:
    • If all of the problems reported are in the snoop.log logfile, you can re-run just snoop (without the SD preview) by using snoop's -d option, for example,

      /usr/sbin/snoop -a -d

      Running snoop this way will not help you resolve problems reported in /var/adm/sw/swagent.log, and will not re-check disk space, so it is important to run snoop once more with -a -o -s (as described above) before you do the actual upgrade.

    Each time you run snoop, new information is appended to the snoop.log logfile. This happens whether you run the program interactively or unattended.

    If you want a fresh log each time you run snoop, you can simply delete the file /var/adm/snoop.log before re-running snoop.

    CAUTION: Do not remove the /var/adm/sw directory; this will remove files that the upgrade and pre-upgrade software depends on.

    If you do accidentally remove /var/adm/sw, run /etc/update to re-install the SD- filesets shown in “Locating and Loading Tools and Documentation” earlier in this chapter.

    See “Handling snoop PROBLEMS”, later in this chapter, for suggestions for solving PROBLEMS that snoop may report.

Modifying HP-UX Clusters

The subsections that follow provide instructions for modifying HP-UX ("DUX") clusters so that they will convert smoothly to NFS Diskless during the upgrade to 10.01.

Removing Series 300/400 Clients and Inactive Clients

Before you get to this section, you should have made provision for your Series 300 and 400 clients, and they should be up and running in the new configuration. See “Providing for Series 300 and 400 Clients ”.

CAUTION: DO NOT PROCEED with this section if Series 300/400 clients are still booted from the Series 700 server.

Now you need to clean up the Series 700 cluster. snoop has identified the Series 300 and 400 clients you need to remove (snoop refers to them as "68K" clients because of their 680n0 processor chips).

Do the following:

  1. Remove Series 300 and 400 clients.

    Use the SAM Remove Cluster Clients screen. Accept the option to remove all client-specific files as well as the clients themselves.

    CAUTION: You must remove Series 300 and 400 clients before you begin upgrading to 10.01. The upgrade software will not do this for you; if it finds Series 300/400 clients, it will exit, leaving the clients intact.
  2. Remove any inactive clients — clients you never intend to boot again from this server.

    Use SAM and remove all the files as with the Series 300 and 400 clients.

    You must also decide what to do about clients snoop has identified as having PROBLEMS. These will be clients that appear in /etc/clusterconf (in entries that have not been "commented out") but do not have a kernel or critical configuration files. snoop refers to these as "invalid" clients. (Clients whose entries have been "commented out" in /etc/clusterconf are also reported as "invalid".)

    • If you intend to boot these "invalid" clients under NFS Diskless on 10.x, fix them as snoop indicates.

      For example, build a kernel if that is what is missing; see the Series 700 9.0 manual Managing Clusters of HP 9000 Computers for more information.

    • If you don't intend to boot them from the 10.x cluster, use SAM to remove them now.

What if You Do Not Clean Up

You may be wondering about the consequences of not doing the work described in the previous subsection. They are as follows:

  • If you do not remove Series 300 and 400 clients:

    • The upgrade will fail, leaving the 9.x system intact. You'll see a message such as:

      Cluster has either S300/400(68K) clients or other
      unknown architecture clients. The UPGRADE process
      does not support heterogeneous upgrade to 10.x.
      User must remove 68K clients and then restart the
      upgrade process.
  • If you do not remove inactive clients:

    • If the client does not have a kernel, all its critical configuration files, and a valid IP address, the conversion will not succeed, and you will see WARNINGS during the upgrade.

      Depending on what parts of the "DUX" configuration are missing, some or all of configuration files that do exist may remain on the server after the upgrade.

      For example:

      • For a client lacking a valid IP address, but otherwise intact, the kernel and all HP-UX configuration files will be saved as regular HP-UX files (that is, not CDFs) on the server under /etc/upgrade/save/client;

      • For a client "commented out" in /etc/clusterconf, or lacking a kernel or critical configuration files ("invalid client") no files will be saved. (A kernel that is a symbolic link to a valid kernel file does not make the client "invalid".)

    • If the client has everything it needs to function under "DUX", then you won't see any WARNINGS about missing files during the upgrade.

      You will see messages indicating that the client is not booted to the cluster (assuming that it isn't; snoop and upgrade refer to such clients as "offline clients"); and the client will show up on the menus of the upgdiskless tool that you will run immediately after the upgrade completes on the server (upgdiskless completes the upgrade and conversion of the clients; see “Upgrading HP-UX ("DUX") Cluster Clients ”).

      Assuming you do not choose to activate the client via upgdiskless, its files will remain on the server under /etc/upgrade/save/client.

      (If you do choose to activate a "valid" client that was not booted to the server during the upgrade, upgdiskless will build a minimally-configured client, using defaults similar to those for a new client created in SAM. See “If Clients Are Not Booted” for details.)

Summary: As you can see, failure to remove Series 300 and 400 clients is a showstopper. The other cases are less serious; the main consequence is that you consume disk space unnecessarily — depending on how much free disk space you have, this may or may not be a serious problem.

Handling Non-"System" CDFs

CDFs are context-dependent files, the files that are "owned" by particular nodes within the global file-system. See chapter 2 of the 9.0 version of Managing Clusters of HP 9000 Computers if you need more explanation.

Some clusters contain only HP-created "system" CDFs (kernels, device files and system configuration files such as /etc/inittab). If this is true of your system, you can skip the next subsection: the upgrade software will convert HP "system" CDFs to regular files and make them available to the nodes that need them on the 10.01 system.

If you have created CDFs, or if you run applications that create CDFs or are themselves stored as CDFs on the system, then you have another task to do.

Saving Non-"System" CDFs

Non-HP CDFs, and CDFs used by HP applications that are not part of the core system, will appear to the 10.x file system as ordinary directories. This probably is not what you want.

NOTE: If the CDFs are in the 9.x "system" directories /bin, /etc, lib, /system, /tmp, or /usr, they will be saved on the server as regular (non-CDF) files, as /directory/CDFname for the server's element and as /directory/!CDFname/element for the other elements.

If you know where they reside in the file system, you can prepare non-"system" CDFs for the upgrade as follows:

  1. Copy each element to a regular HP-UX file.

    Make the names relate in some obvious way to the nodes that own the files. For example, you might copy /apps/myapp+/node1 to something like /10.copies/myapp.node1.

  2. Make a backup of the regular files.

    Delete the copies on disk if you like.

  3. If you have any directories that are CDFs (apart from those created by HP-UX when you configured the cluster) repeat the above steps for the directories.

    Copy the elements to a regular directory and back the regular directories up to tape.

  4. Just before upgrading to 10.01, and after you have done a backup with the -Hoption, remove all non-"system" CDFs from the system.

    The -H to backup commands such as fbackup(1M) backs up all elements of CDFs.

    See “Backing Up Your System ” at the end of this chapter.

Reclaiming Device Swap

If you see a WARNING in the snoop logfile that you have more device swap on the server than you are likely to need under NFSD, you should consider doing two things:

  1. Create file systems on disks that are currently dedicated to device swap.

    Plan to do this during the period when the cluster is down for the upgrade, either just before or just after you upgrade the server, but before trying to boot the clients under NFS Diskless.

  2. Create a file which will tell upgrade to reclaim device swap at the end of file systems on the server's disk(s) and add it back to the file system.

    See “Creating the Swap Reclamation File ”, later in this section.

    NOTE: The default for NFS Diskless clients is to swap to the server's file system space. You cannot configure device swap to any remote disk, including the server's. Make sure you understand how upgrade and upgdiskless (the program you will run right after the upgrade to re-activate the clients) convert various "DUX" swap configurations to NFS Diskless swap: see the subsection on "Swap" under “Upgrading a "DUX" Cluster”.
Creating the Swap Reclamation File

Create the swap reclamation file as a plain ASCII file, using your favorite editor.

It must contain entries in the form:

character_device_filename  Mb_to_reclaim

For example:

/dev/rdsk/c201d0s0          80
/dev/rdsk/c201d1s0 155
/dev/rdsk/c201d2s0 107
NOTE: You must supply the character (or "raw") device-file name, not the block device file name. By default, character device files are stored in the directory /dev/rdsk.

Entries must be one line per disk, fields delimited by one or more space characters (any combination of spaces and tabs).

Comment lines preceded by the pound character (#) are allowed.

Save the file in following path:

/etc/upgrade/save/server/ttools/AllSwap.d/reclaim_swap

where server is the name of the server as returned by uname -a. snoop has already created the directories in this path.

When you run snoop again, snoop will check for the file and check the validity of the entries. You will see CAUTIONS if there are errors in the file; if the file does not exist you'll see the same CAUTION as before about excessive device swap. If the file exists and has no errors, you'll see a NOTE that snoop found a valid reclamation file, and a list of its contents.

"Unsecuring" a Trusted System

If the cluster is a Trusted System, you will need to "unsecure" it before you can upgrade it. You will probably want to wait to do this until just before running the upgrade.

To "unsecure" the cluster, enter the following command on the server:

/etc/tsconvert -r

After the upgrade to 10.01 is complete, and you have run upgdiskless to add back the clients, you will need to "resecure" the cluster; see “Resecuring a Trusted System Cluster ”.

Handling snoop PROBLEMS

This section provides suggestions for correcting PROBLEMS you may find in the logfile /var/adm/sw/snoop.log.

If You Have Less than 16 Mb of Memory

You cannot upgrade a system that has less than 16Mb of memory. This includes cluster clients.

Either:

  • Order more memory; or

  • Do not upgrade this system.

If you decide not to upgrade the system, remember that you can still NFS mount file systems to and from other 10.x systems in the network, so long as you do not try to mount system directories. See “Moving Files ”, and “Operating 9.x and 10.x Systems Together ”, for more information.

If You Have an 8Mb Memory Card in Slot One (Series 800)

If you got a warning from snoop about this, it does not necessarily mean that you have this configuration, only that you could have.

To find out for sure, do the following:

  1. Reboot the system and interrupt the boot by pressing any key.

  2. Answer n when prompted to boot from primary path.

  3. Answer n when prompted to boot from alternate path.

  4. Answer y when asked if you want to interact with ISL.

  5. At the ISL prompt, enter iomap.

  6. You'll be asked if you want to modify program parameters; answer n.

    Now you'll see a display at the end of which the memory cards are listed (under Memory Controller).

    If the first card listed is an 8Mb card, you cannot upgrade this system as it stands.

    • If you can swap another memory card into this slot, do so.

    • If this is not possible, or you are not comfortable doing it, contact your HP representative for help.

  7. iomap asks you if you want to return to ISL. Answer y.

  8. Boot the system:

    hpux

If There Are "PROBLEMS" on an HP-UX Cluster Server

  • If snoop reports that you still have Series 300/400 clients, you must remove them before you can upgrade the cluster. See “Removing Series 300/400 Clients and Inactive Clients ”, earlier in this chapter, and “Providing for Series 300 and 400 Clients ” for details.

  • If snoop reports that some of your clients were not booted to the server, you should, if possible, boot these clients to the server and re-run snoop.

    This is the only way to be certain of catching all potential problems before you do the actual upgrade. If the clients are not booted to the server during the upgrade itself, they will, at best, lose much of the benefit of the conversion (see “If Clients Are Not Booted”); at worst (if there were problems that have gone undetected because the clients were not booted during any run of snoop) the clients may not be converted at all.

    If you never intend to boot these clients from this server under 10.x, you should remove them from the cluster before you do the upgrade; see “Removing Series 300/400 Clients and Inactive Clients ”, earlier in this chapter.

  • If snoop reports that it was unable to convert local swap defined in the difle (the client's kernel-configuration file) to a swap entry in /etc/checklist, you will need to create the entry in /etc/checklist yourself.

    Modify /etc/checklist+/client, where client is the name of the client reported by snoop. The HP-UX System Administration Tasks manual explains how to create swap entries in /etc/checklist.

    NOTE: You also need to create a swap entry in /etc/checklist+/client if you have activated local swap only from the command line (with a swapon command) — that is, if there is no entry in the client's dfile or checklist file to make the local swap "permanent".
  • If snoop reports that this is a Trusted System, you need to "unsecure" the system before the upgrade can proceed; see “"Unsecuring" a Trusted System ” earlier in this chapter.

If You Do Not Have Enough Disk Space

snoop's disk space analysis errs on the conservative side; that is, it's possible snoop may tell you you don't have enough space when in fact you have just enough, but it is unwise to gamble unless this is a test or non-production system and you can afford the time to experiment.

NOTE: An HP-UX ("DUX") cluster server is more likely to fail disk space analysis than other systems, because the algorithm is allowing for file-system swap to the server's disk space. See “Swap ” for a discussion.

Remedies

Remedies depend on the nature of the problem snoop reports:

  1. If you don't have as much total free space on your system as snoop says you need, you may have to add a disk or disks before you can upgrade.

    But you may have more software on your system than you need, or you may have told snoop (by "marking" additional bundles or products) that you intend to add 10.01 features that you don't actually need.

  2. If you have enough space in total, but not on the root disk, you may be able to redistribute the disk space.

    • On a Series 800:

      • If the problem is with / (root), do one of the following:

        • If using LVM, move the root volume to a larger logical volume.

          This requires an unused disk that will become the new bootable device. Use the procedure that begins on page 8-121 of the 9.0 version of the Series 800 HP-UX System Administration Tasks manual.

        • If using hard partitions, move directories under / (for example /tmp) to another partition.

          NOTE: This will not work for directories such as /etc, /lib and /dev.
        • If using hard partitions, move the root partition.

          See Chapter 6, "Managing the File System," in the 9.0 version of the HP-UX System Administration Tasks manual for details.

      • If the problem is with /usr, do one of the following:

        • If using LVM, extend the /usr volume.

          See Chapter 8, "Managing Logical Volumes," in the 9.0 version of the HP-UX System Administration Tasks manual for details. This procedure cannot be performed via SAM.

        • If using hard partitions, move /usr to a larger partition.

        • Move /usr to another disk.

          See Chapter 6, "Managing the File System," in the 9.0 version of the HP-UX System Administration Tasks manual for details.

    • On a Series 700:

      • Move /usr to another disk.

        See Chapter 6, "Managing the File System," in the 9.0 version of the HP-UX System Administration Tasks manual for details.

Removing Unneeded Filesets

When you choose Match What Target Has during the upgrade, SD will load all software that corresponds to what is on your 9.x system. If snoop reports you are short of disk space, make sure, before you do the upgrade, that you do not have filesets on your system that you don't need.

Likely candidates for removal are Native Language Support (NLS) and (on Series 800) graphics filesets. See “Series 800 Small System ”, later in this chapter, for more information.

If you have the HP-UX "manpages" in hardcopy or on CDROM, you may want to remove the online version.

Use the rmfn utility, described in chapter 2 of the 9.0 HP-UX System Administration Tasks manual, to remove 9.x filesets you are not using and whose functionality you will not need on 10.x.

NOTE: If you have more than one user licence on your system (for example, a 2-user licence and an 8-user licence), remove the licence you do not use before you upgrade. Check the /system and /etc/filesets directories for licence files such as 02-USER and 08-USER.
Series 800 Small System

By "small system" we mean a system that probably has these characteristics:

  • 16-32 Mb of RAM.

  • Single hard disk of 500Mb or less.

  • Disk space is at a premium.

  • System does not use graphics (X11, Vue, etc.) or Native Language Support (NLS).

The last two bullets in the above list are crucial; read the discussion below, and “Filesets To Remove ”, later in this chapter, before you make a decision.

HP-UX 10.01 includes bundles (groupings of products and filesets) for the Series 800 that include the run-time system, but exclude NLS and graphics filesets that many Series 800s do not need. If you intend to install one of these bundles on your Series 800, do the following when you upgrade to 10.01:

  • Choose Match What Target Has

  • Proceed with the upgrade.

  • Once the initial upgrade has succeeded, run SD again and this time select the "non-graphics" bundle for the human language that the system will use.

    See the next subsection, “Non-Graphics Bundles ”.

    These bundles include features such as JFS that are new for 10.01.

  • Select any other bundles you have ordered in addition to HP-UX.

See “Loading New 10.01 Features and HP Applications ” for more information.

This can save you more than 50Mb of disk space, but only if you first remove the graphics and NLS filesets listed under “Filesets To Remove ” below.

Non-Graphics Bundles

The following are the names of the non-graphics bundles as they appear on SD's Software Selection screen when you set software view to All bundles. Choose the bundle that corresponds to the human language your system will use; for example, if the system's users are primarily English-speaking, you would choose HPUXEngRT800.

Bundles                     Descriptions

HPUXDanRT800 Danish non-Graphics Runtime HP-UX Environment
HPUXDutRT800 Dutch non-Graphics Runtime HP-UX Environment
HPUXEngRT800 English non-Graphics Runtime HP-UX Environment
HPUXFinRT800 Finnish non-Graphics Runtime HP-UX Environment
HPUXFreRT800 French non-Graphics Runtime HP-UX Environment
HPUXGerRT800 German non-Graphics Runtime HP-UX Environment
HPUXItaRT800 Italian non-Graphics Runtime HP-UX Environment
HPUXJpnRT800 Japanese non-Graphics Runtime
HPUXKorRT800 Korean non-Graphics Runtime
HPUXNorRT800 Norwegian non-Graphics Runtime HP-UX Environment
HPUXSChRT800 Simp.Chinese non-Graphics Runtime
HPUXSpaRT800 Spanish non-Graphics Runtime HP-UX Environment
HPUXSweRT800 Swedish non-Graphics Runtime HP-UX Environment
HPUXTChRT800 Trad. Chinese non-Graphics Runtime
Filesets To Remove

The following lists show what filesets you can remove from your 9.x system if you do not use graphics, or Native Language Support, or both.

Graphics Filesets
=================

AUDIO
AUDIO-MAN
BMS
CAPTURE
IMAGING-RUN
IMAGING-SHLIBS
IMAGNG-RUN-MAN
SHPRNT-HELP
SHPRNT-PCL
SHPRNT-INSTALL
SHPRNT-PCONFIG
SHPRNT-CLI
SHPRNT-NLS
SHPRNT-SRV
X11-FONTA
X11-FONTB
X11-FONTC
X11-FONTSRV
X11-RUN
X11-RUN-HELP
X11-RUN-MAN
X11R4-SHLIBS
X11R5-SHLIBS
NLS Filesets
============

Language Fileset Name
-------- ------------

Arabic ARABIC
ARABICW
Bulgarian BULGARIAN
Czech CZECH
Chinese-Simplified CHINESES
CHINESES-CAT
CHINESES-X11
Chinese-Traditional CHINESET
CHINESET-CAT
CHINESET-X11
Danish DANISH
Dutch DUTCH
Finnish FINNISH
French CFRENCH
CFRE-X11-HELP
FRENCH
FRENCH-CAT
FRENCH-X11
FRE-X11-HELP
German GERMAN
GERMAN-CAT
GERMAN-X11
GER-X11-HELP
Greek GREEK
Hebrew HEBREW
Hungarian HUNGARIAN
Icelandic ICELANDIC
Italian ITALIAN
ITALIAN-CAT
ITALIAN-X11
Language            Fileset Name
-------- ------------

Japanese JAPANESE
JAPANESE-CAT
JAPANESE-MAN
JAPANESE-X11
JEUC-CAT
JEUC-MAN
JEUC-X11
JEUC-X11-HELP
JPN-X11-HELP
Korean KOREAN
KOREAN-CAT
KOREAN-X11
KOR-X11-HELP
Norwegian NORWEGIAN
Polish POLISH
Portuguese PORTUGUESE
Rumanian RUMANIAN
Russian RUSSIAN
Serbocroatian SERBOCROATIAN
Slovakian SLOVAKIAN
Slovene SLOVENE
Spanish SPANISH
SPANISH-CAT
SPANISH-X11
SPA-X11-HELP
Swedish SWEDISH
Thai THAI
Turkish TURKISH
NOTE: You can also remove the MKFONTDIR fileset if you use neither graphics nor Asian languages: Japanese, Simplified Chinese, Traditional Chinese or Korean.

You must remove the filesets listed above in order to benefit from choosing a "non-graphics" bundle as described in the previous sub-section, “Series 800 Small System ”.

Proceed with caution when removing filesets.

  • Do not remove an NLS fileset, and do not choose a "non-graphics" bundle, if you use or develop applications in the language in question.

    (But you should still be able to remove the filesets for the languages you don't use).

  • Do not remove the graphics filesets, and do not choose a "non-graphics" bundle, if you use graphics or multimedia applications such as screen capture or SharedPrint, if your Series 800 runs Xterminals or acts as an HPVue or X server, or if you use NetLS.

  • Always use rmfn to remove filesets; this ensures that you do not remove filesets that are required by other filesets that you need.

What To Do Next

The sections that follow outline other tasks you may need to do before upgrading to 10.01. Read the sections, do whatever tasks apply, then run snoop again to check that you have solved all PROBLEMS reported in the logfile.

Printable version
Privacy statement Using this site means you accept its terms Feedback to webmaster
© 1983-1995 Hewlett-Packard Development Company, L.P.