Discless system configuration tasks - technical

Discless System Configuration Tasks

GOING FROM A GROUP of stand-alone machines to a clustered environmental is not a particularly difficult task, but because of the large number of steps required to configure the system, an automated tool called reconfig is provided with the HP-UX discless system to simplify the process. Reconfig enables the system administrator to set up the cluster server node and add or delete cluster nodes (cnodes) as necessary.

Reconfig was originally developed for the HP 9000 Series 200 and 300 computers' HP-UX 5.1 operating system. The tool contains a collection of monotonous and terse system administration tasks within a user-friendly menu-driven environment. basic tasks such as setting up user access to the system and reconfiguring kernels can be easily accomplished. With the advant of discless workstations in a clustered environment, changes were made to enhance the original reconfig tool.

Cluster Setup

For creating a cluster configuration, the minimum system includes an HP 9000 Model 350 for the root server with at least 8, bytes of RAM, at least a 130-Mbyte disc drive, and the HP-UX 6.0 operating system (or later). The setup process begins by running /etc/reconfig, and when the main menu appears selecting the option Cluster Configuration. This selection will bring up the menu shown in Fig. 1, which shows the four values required to set up a cluster server: the server node name, the link level LAN address, the internet address, and the number of cluster server processes (CSPs).

Cluster Node Name. The server's cnode name is the system's hostname and it is used to identify the server cnode within the cluster. All discless cnodes refer to the root server by this name.

LAN Card's Link Level Address. Each LAN interface card has a unique link level address. This value is set by the factory and cannot be changed. Reconfig will dispaly the address for each LAN attached to the system. If there is only one LAN card on the system its address is used by default; otherwise, one of the available cards must be selected using the NEXT, PREVIOUS, and SELECT softkeys.

NS-ARPA Internet Address. The internet address enables communication with other networks and uniquely indentifies the server within a network. The internet address is not required for discless interaction, but provides a minimal NS-ARPA networking capability within the cluster to handle remote process execution for system administrative tasks. If a value is automatically displayed, that value is the internet address associated with the cnode name that already exists in the system's /etc/hosts file.

Cluster Server Process (CSP). The CSP is a special process that is used to handle interprocess communication in a discless environment. Except for one limited CSP (LCSP) which exists on each discless cnode, all the other CSPs exist on the server. The default value is 4 and the amount entered will be the minimal number of CSPs running at all times. If more CSPs are needed they will be created automatically. For an in-depth discussion of CSPs see the article "The Design of Network Functions For Discless Clusters" on page 20.

When all the entries in the menu have been entered reconfig will tell the user what it is about to do to build the system and then ask if the user wants to continue. A yes will cause reconfig to begin configuration. Reconfig performs five steps in transforming the stand-alone system to a clustered environment. The steps are done in a particular order, so if an error occurs during the process, the user can correct the problem and then reexecute reconfig from where it left off. Generally, each step in the process will complete without error unless certain required files and/or directories are missing. The activities that take place at each step are as follows:

* Context dependent files (designated by a + appended to the file name) are created for the cluster and root server based upon a predetermined system model (see Fig. 2). Context dependent files (CDF) are used to describe the various attributes (e.g., machine type, coprocessors) of a particular cnode. For more information on CDFs see the article "A Discless HP-UX File System" on page 10.

* A fully loaded root server kernel is built. The directory /hp-ux is turned into a CDF and the new version of the kernel resides in /hp-ux+/<cnode_name>. Cnode_name is the name entered earlier for the cluster node name.

* The NS-ARPA files for remote process execution are set up and an entry is made for the root server for the following files: /etc/hosts,/etc/hosts.equiv, and $HOME/.rhosts (root's home directory).

* The cluster configuration file /etc/clusterconf is created and the following information is entered in the file for the root server: the root server's cnode name, the server's link level LAN address, and the number of CSPs to start at boot.

* The rc file, which initiates the boot service for the discless cnodes, is modified to state which LAN device file to use if the default LAN device file is inappropriate.

* Reboot system.

Adding and Deleting Cnodes

Once the root server of the cluster has been set up, discless cnodes can be added or deleted at will by running /etc/reconfig and selecting the Cluster Configuration option from the main menu. If the root server has already been set up (e.g., /etc/clusterconf exists), reconfig will present two menu choices for adding or deleting discless cnodes.

Adding a Cnode

The input required for adding a cluster node is similar to that required for initial cluster setup: the cnode name, an internet address, and the link level address of the cnode's LAN card. Each discless cnode always runs exactly one limited cluster server process (LCSP) so there is no need to prompt for the number of cluster server processes. The process for adding a cnode is similar to that for setting up the clustered environment on the root server. The four steps are as follows:

* Context dependent files are created for the new discless cnode based on the system model for client cnodes (see Fig. 3).

* A minimally loaded discless cnode kernel is built. The directories /hp/ux+/<cnode_name> and /etc/conf/dfile+/<cnode_name> are created.

* NS-ARPA files for remote process execution are set up for the discless cnode. The file /etc/hosts, /etc/hosts.equiv, and $HOME/.rhosts (root's home directory) are modified to include the new cnode.

* The cluster configuration file (clusterconf) is modified to include an entry for the discless cnode. The entry includes the new cnode's name and its link level LAN address.

Removing a Discless Cnode

Only the discless cnodes can be removed with reconfig. All that is required to remove a discless cnode with reconfig is the cnode name. The menu shown in Fig. 4 is used to select the cnode to be removed. There is an option to remove or not to remove all CDFs associated with the cnode. Unless there is a good reason for leaving the CDF elements around, the CDFs should be removed when the discless cnode is removed. The cnode removal process involves the following steps:

* Remove the ability to do remote process execution by deleting the entries for the cnode from the NS-ARPA files /etc/hosts.equiv, and $HOME/.rhosts. The cnode name and its associated internet address remain in the file /etc/hosts for later use.

* Remove the entry in the cluster configuration file (/etc/clusterconf) for the deleted cnode.

* If requested, remove all context dependent file elements of the form: <file>+/<cnode_name>.

Conclusion

The Reconfig tool provides features that make the tasks of setting up and maintaining an HP-UX discless cluster much easier. In addition, the time required for reconfiguration is much lower with Reconfig than it would be to administer each system individually. This one of the advantages of the HP-UX discless system.

Acknowledgments

Special thanks go to Stuart Bobb and Dave Grindeland for their usability testing efforts, and to Paul Christofanelli and Paul Van Farowe for their invaluable NS-ARPA networking assistance.

COPYRIGHT 1988 Hewlett Packard Company
COPYRIGHT 2004 Gale Group