Converting HP-UX ("DUX") Clusters

HP-UX ("DUX") clusters are no longer supported as of HP-UX 10.0.

The package that includes this manual provides tools that convert clusters of Series 700 computers from HP's proprietary technology, "DUX", to the new technology introduced at 10.01, which is based on NFS Diskless.




	NOTE: HP recommends that you use these tools to upgrade Series 700 "DUX" clusters. This is the only method HP supports for upgrading a "DUX" cluster to 10.x. No Series 300 or 400 computer can be a member of an NFS Diskless cluster, nor can these computers be upgraded "standalone" to HP-UX 10.x. See “Providing for Series 300 and 400 Clients ”, later in this chapter.

Upgrading a "DUX" Cluster

As with a standalone system, you must upgrade a cluster to 10.01 before you can update it 10.10.

In outline, the steps for upgrading a "DUX" cluster from 9.x are as follows. You'll find detailed procedures later in this chapter, and in chapters 4, 6 and 7, as noted below.

Remove all Series 300 and 400 clients from the cluster.
See “Providing for Series 300 and 400 Clients ”, later in this chapter.
Remove all inactive clients and client "fragments".
"Inactive clients" means clients that you never intend to boot to the new NFS Diskless cluster; "fragments" are kernels or device or configuration files for clients that are no longer part of the cluster.
See “Removing Series 300/400 Clients and Inactive Clients ”.
Run the analysis tools on the cluster server.
See “Analyzing Your System ”.
Upgrade the server to 10.01.
See Chapter 6 “Upgrading Your System from 9.x to 10.01 ”.
Upgrade the clients.
See “Upgrading HP-UX ("DUX") Cluster Clients ”.
Upgrade applications, and update the cluster to 10.10 if you so decide.
See Chapter 7 “After the Upgrade ”.

What Does the Conversion Do for You?

The "DUX" upgrade tools convert a running 9.x HP-UX cluster to a 10.01 NFS Diskless cluster. As far as possible, the conversion happens automatically, restoring "DUX" functions (or their nearest equivalents) on the NFS Diskless clients, and requiring you to intervene only when decisions have to be made.

The tools expect all cluster clients to be booted to the cluster server at the time of the upgrade (and of the pre-upgrade analysis). See “Running snoop ” and Chapter 6 “Upgrading Your System from 9.x to 10.01 ”. The table that follows shows what "DUX" functions will be restored on these clients when they come up on 10.x under NFS Diskless.

For information on what happens to clients that are not booted to the server, see “If Clients Are Not Booted”, later in this chapter.

"DUX"           Restored        "DUX"           Restored
Function        on 10.01?       Function        on 10.01?
===========================================================
 
FILE SYSTEMS:                   SPOOLING:
- on server's                   - to printers
  disks:           no*            attached to
- on client's                     server:          yes
  own disks:       yes          - to printers
- on another                      attached to
  client's                        this client:     yes
  disks:           no**         - to printers
                                  attached to
SWAP:                             another
- to root                         client:          yes
  server:          yes          - to network
- to local                        printers:        yes
  disk:            yes
- to swap                       SYSTEM CONFIG
  server                        INFO:
  (device swap                  - networking       yes
  to another                    - critical
  client's                        config files
  disk):           no             (inittab,
- to file                         fstab..)         yes
  server
  (f/s swap                     DEVICE FILE
  to another                    INFO:              yes
  client's
  disk):           yes          EMAIL:             yes
 
BOOT FROM                       CLUSTER-WIDE
SERVER:            yes          LOGINS:            yes
 
KERNEL:            yes          APPLICATIONS       no~
 
*  =  HP-UX "system" directories only; use SAM to export others
** =  Use SAM to NFS-mount these file systems
~  =  See "Applications" later in this chapter

The subsections that follow provide more information about the conversion.

File Systems

In a 9.x "DUX" cluster, all file systems on the server are automatically visible to the clients, and a file system resident on a client's local disk is visible to the server and all the other client.

This is not the case under NFS Diskless. Directories that comprise the HP-UX operating system, and directories such as /export/private_roots/client that contain the client's configuration files (and, by default, swap files) are automatically exported, but other directories on the server's and clients' disks are not.

The following directories are automatically exported from the server and mounted on all clients:
/sbin /usr /opt/* if optional products are configured for diskless /home if policies specify shared home dirs /var/mail if policies specify shared mail
For information on "shared" versus "private" cluster policies, see the NFS Diskless Concepts and Administration white paper; “Locating and Loading Tools and Documentation” explains how to get it off the tape or disk that is packaged with this manual.
Other directories are not automatically exported and mounted; you need to do this explicitly. SAM, the menu-driven System Administration Manager simplifies the task. See “Procedure for Upgrading Clients ” for more information.

See Appendix B “NFS Diskless Questions and Answers” for more information on managing file systems and disk space in an NFS Diskless cluster.

Swap

By default, NFS Diskless clients swap to their /paging directory, which is in the server's disk space (/paging on a client is /export/private_roots/client/paging on the server). But, as far as possible, the 9.x-to-10.01 upgrade preserves the swapping arrangements used by your 9.x "DUX" cluster.

For clients that are booted to the server during the upgrade, the upgrade program converts DUX swap configurations to NFSD as follows:

Table 2-1 Title not available (Swap )

"DUX" Configuration	Converted to NFSD Configuration
Device swap to server	File-system swap to server
Device swap to local disk	Device swap to local disk
Device swap to another client	File-system swap to server

In addition, file-system swap configured under "DUX" will be restored under NFS Diskless.




	CAUTION: Do not configure an NFS Diskless client to swap only to another client's file system. Configure primary swap to the server or a local disk; otherwise the client may fail to boot.

Kernel-Configured Swap. Under 9.x, swap to a client's local disk can be configured via a statement in the client's dfile; on NFS Diskless clients it cannot.

If the client is booted to the server during the upgrade, the upgrade process will convert local device swap configured in the dfile to local device swap configured in /etc/fstab (the 10.x equivalent of /etc/checklist).




	NOTE: If you have configured local swap both in the `dfile` and in `/etc/checklist`, the upgrade process will choose the `checklist` version (whereas 9.x HP-UX uses the `dfile` version). If you have configured swap by means of a command-line `swapon` statement only, and have not put a swap entry in either the `dfile` or `/etc/checklist`, this client will get default swap (file-system swap to the server).

Too Much Device Swap? By default, NFS Diskless clients swap to the server's file system, whereas under "DUX" they swap by default to the server's device swap space (device swap refers either to a swap area at the end of a file system, or to a disk dedicated to swap that has no file systems on it).

In fact, NFS Diskless clients cannot use device swap on the server, or any other remote system, but only on their own local disks. (NFS Diskless clients can use NFS file-system swap on a remote system, whether or not that system is part of the cluster, and whether or not it is running HP-UX 10.x.)

Because of these differences, you may have too much device swap configured on the server's disks, and not enough room for file-system swap.

For 9.x "DUX" clusters, HP recommended 30Mb device swap per client as a rule of thumb. Under NFS Diskless, HP recommends a minimum of 44Mb file-system space per client for all purposes, including kernel, boot files and file-system swap. By default, this space is used under /export on the server (but you can allocate the space elsewhere and link it to directories under /export, as explained in Appendix B “NFS Diskless Questions and Answers”).

You should probably allow for this much space whether or not clients are configured to swap to their local disks, because all clients swap to the server while they are booting.




	NOTE: When you run the `snoop` analysis tool, described in Chapter 4 “Pre-Upgrade Tasks for All 9.x Systems”, and `upgrade`, described in Chapter 6 “Upgrading Your System from 9.x to 10.01 ”, the tools will do a disk-space analysis on the basis of the software you select. In the case of a cluster server, this analysis assumes that all the clients will be swapping to the server's disk space (always true when they are booting). This conservative analysis may show a surprising increase over 9.x in the amount of disk space required, but it ensures that all the clients will be able to boot at the same time (for example, when you first bring them up as NFS Diskless clients).

Remedies:

The snoop and upgrade tools allow you to reclaim device swap that is at the end of the file system space on the server's disks.

snoop will warn you if you appear to have more device swap on the server than you will need, and at that point, before you run upgrade, you can create a file that will be read during the upgrade process; upgrade will reclaim the amount of swap you specify, on the disks you specify, and add it back to the file system.




	CAUTION: This will be your only chance to reclaim this space without doing a full backup of the data on the disk. See “Reclaiming Device Swap” and “Procedure for Upgrading Clients ” for more information.

You can use ordinary HP-UX commands to create a file system on a disk that was formerly dedicated to device swap.
You should probably do this either just before or just after upgrading the server to 10.01, so as to keep your "DUX" cluster up and running as long as possible, but in time to make server swap space available before you boot the clients under NFS Diskless.

See Appendix B “NFS Diskless Questions and Answers” for more information on managing disk space and swap in an NFS Diskless cluster.

Applications

Not all HP and third-party applications are certified to run on NFS Diskless clusters. Contact the suppliers of critical applications before you commit to upgrading a cluster to 10.x.

If Clients Are Not Booted

The tools will provide defaults for clients that are not booted to the cluster server while it is being upgraded, but, as the table below shows, you will need to do more work to recapture the 9.x configuration of these clients; they will look to the new NFS Diskless cluster much like new clients, lacking information about peripherals, users, etc. (The snoop analysis tool, discussed in Chapter 4 “Pre-Upgrade Tasks for All 9.x Systems”, refers to such clients as "offline" clients.)

"DUX"           Restored        "DUX"           Restored
Function        on 10.01?       Function        on 10.01?
===========================================================
 
FILE SYSTEMS:                   SPOOLING:
- on server's                   - to printers
  disks:           no*            attached to
- on client's                     server:          yes
  own disks:       no%          - to printers
- on another                      attached to
  client's                        this client:     no
  disks:           no**        - to printers
                                  attached to
SWAP:                             another
- to root                         client:          no
  server:          yes          - to network
- to local                        printers:        no
  disk:            no%
- to swap                       SYSTEM CONFIG
  server:          no%          INFO:
- to file                       - networking       no%
  server:          no%          - critical
                                  config files
BOOT FROM                         (inittab,
SERVER:            yes            fstab..)         no%
 
KERNEL:            no%          DEVICE FILE
                                INFO:              no%
CLUSTER-WIDE
LOGINS:            yes          APPLICATIONS       no~
 
EMAIL:             yes
 
*  =  HP-UX "system" directories only; use SAM to export others
** =  Use SAM to NFS-mount these file systems
%  =  Defaults similar to new NFS Diskless clients;
      all attached peripherals must be reconfigured
~ =   See "Applications" earlier in this chapter

Clients with "Dual Personalities"

HP-UX 10.x releases are designed to work well with previous HP-UX releases. This means, for example, that if a client is a member of two "DUX" clusters, you can upgrade one cluster to 10.x, boot the client to the 10.x cluster, and from then on boot the client to either server, whether the servers are running 8.x and 10.x, 9.x and 10.x, or 10.01 and 10.10.

Similarly, 10.x clients will be able to boot standalone from a local disk containing any supported O/S.

Of course, if you reconfigure the client's disks, you may destroy or compromise the client's "other personality". For example, if you configure a 10.x client's local disk for LVM, then boot the client to a 9.x cluster, the 9.x client will not be able to use the data on the disk, and must not swap to it.

Managing Root's Home Directory

In HP-UX "DUX" clusters, all directories are visible on all systems in the cluster, so the root user's home directory is automatically available on all cluster nodes. This is not true for NFSD: the root user's home directory is not automatically shared on the clients (whether or not root's home directory is /).

You could force the root user's home directory to be shared, for example by making it a subdirectory of a directory that is shared automatically, such /home, but HP recommends the following:

Make root's home directory / (as it is by default).
Never try to share the / directory.
Limit the files in this directory to the essential "dot" files (.profile, etc.)
Put any scripts or applications you want to reserve for the root user in some other directory in the root user's path, and export the directory to the clients.
See “Procedure for Upgrading Clients ” for more information on sharing directories.

Providing for Series 300 and 400 Clients

Because 10.x does not support Series 300 and 400 clients, you must remove any such clients from a Series 700 cluster before you upgrade the server to any 10.x release. Before you do that, though, you will need to make some other provision for these clients. (The snoop tool, discussed in Chapter 4 “Pre-Upgrade Tasks for All 9.x Systems”, refers to these as "68K clients", as shorthand for their 680n0 processor architecture.)

Here are some of the options:

Convert the clients to standalone machines.
Convert one client to be a Series 300/400 cluster server.
Upgrade clients to Series 700s.
Convert clients to Xterminals.

Each of these options is discussed in the bullets that follow.

Convert the clients to standalone machines.
If the clients have at least one disk each, you can install HP-UX onto each client. Follow directions in the Series 300/400 version of Installing and Updating HP-UX 9.0.
To avoid re-certifying applications, install a version of HP-UX that is compatible with the version currently installed on the Series 700 server (for example, if the server is running 9.03, install the Series 300/400 version of 9.0 or 9.03 on the former clients).
The Series 300/400 systems' disks do not each need to have a copy of all the non-system files stored on the Series 700's disks; instead, they can NFS-mount the directories they need from the Series 700.
Follow the directions for doing NFS mounts in chapter 6, "Managing the File System," of the 9.0 Series 300/400 version of the System Administration Tasks manual (HP part number B1864-90010).
If you choose this option, you may need to buy disks, and you will lose the advantage of "single-point administration."
Convert one client to be a Series 300/400 cluster server.
A Series 300 or 400 cluster server needs at least 16 Mb of memory and at least a 420 Mb disk — enough disk space for HP-UX 9.0, and an additional 30 Mb per client for swap, unless the clients can swap to their own or another client's disk.
See the 9.0 Series 300/400 version of Managing Clusters of HP 9000 Computers (HP part number B1864-9015) for instructions on configuring a cluster server and details of hardware requirements.
Once you have configured the server, add the other Series 300/400 machines as clients, and NFS mount non-system directories from the Series 700 server (see previous bullet).
If you choose this option, you may need to buy more memory or a bigger disk for the Series 300/400 server, and you will have to manage two clusters where there was only one before, but this may be the least expensive and most efficient of the options.
Upgrade clients to Series 700s.
If all of your clients are Series 400s, you may be able to buy board upgrades to convert them to Series 700s. Board upgrades are not available for all models of Series 400; check with your HP Sales Representative (SR).
If it turns out that you can convert all of the Series 400 clients to 700s, do the following to upgrade the cluster to 10.01:
- Remove the Series 400 clients from the cluster.
- Do the hardware upgrade to convert them to 700s.
- Upgrade the cluster to HP-UX 10.01, converting it to NFS Diskless using the special tools provided by HP.
- Upgrade the cluster to 10.10 if you so decide.
- Add back the clients that you converted to 700s.
(Alternatively, once you've done the hardware conversion, you can add the clients back as Series 700 "DUX" clients under 9.x, then upgrade and convert the cluster to NFS Diskless.)
This option leaves your cluster intact and frees up disk space (you now only need the Series 700 version of HP-UX), but is not available for Series 300s and only for certain Series 400 models.
Convert Series 300 and 400 clients to Xterminals.
Xterminals are graphics-display terminals; if your clients are truly running "diskless" (without any disks of their own) and are not supporting local printers or other peripherals, then in effect you are already using them as terminals, and converting them to Xterminals may be your best option.
To convert Series 300 and 400 clients to Xterminals, you will need the XTERM300 product which is part of HP-UX 9.10 (XTERM300 cannot be ordered separately); consult your HP Sales Representative for more information.
For more information on this option, see the section "Using the Series 300/400 Xterminal Product" in chapter 5 of the HP-UX 9.x/10.x Interoperability Guide.

Summary of "DUX" Issues

HP-UX "DUX" clusters are no longer supported as of 10.0.
10.01 introduces a new kind of HP-UX cluster, HP's much enhanced version of NFS Diskless.
The package of which this manual is a part includes special tools to upgrade and convert 9.x "DUX" clusters of Series 700 computers to 10.01 NFS Diskless clusters.
No Series 300 or 400 system can be a member of an NFS Diskless cluster.
You will have to make other provisions for these clients; see “Providing for Series 300 and 400 Clients ” earlier in this chapter. You must remove these clients from the Series 700 cluster before you run the upgrade.
Not all HP and third-party applications will work correctly on all members of a cluster; if in doubt, contact the supplier of the application.