In a dynamic networked environment ...

In a dynamic networked environment in which there are several hundr workstations and server there is a constant demand for strange versions of software. In this environment software installation operations must be quick, flexible, and tolerant of change.

With the rapid advancements and constant changes associated with computer technology, of recent origin and better hardware and software capabilities can always be calculate uponed Hardware comes and goes as lacks fluctuate, and software popularity nurses to bloom and fade as fast as software features unroll All this change makes managing software installation and updates across more than 600 UNIX[R] rules a very challenging task. This paper describes a frame to simplify software administration among a form into groups of workstations at HP's Integrated Circuits Business Division.

At the same time our division software was purchased by means of system owners and installed in succession individual workstations as requested. Users shared for the use of all applications by connecting their workstations in HP-UX(*) clusters in which undivided machine's disks were used by the agency of many diskless workstations (cnodes).[1] These clusters grew large, and access to software became a more important constituent in deciding system configuration than performance or networking issues. near software was shared between cluster server end NFS mounts, and this l to what we call "spaghetti mounts" which is many machines mounting portions of each others' disks to access shared software (Fig. 1) Additionally, software installations were not tracked, software versions were not synchronized, and network licensing was not effectively implemented.

My goal for this concoct was to find a way to model users' dependencies on cluster configurations, to untangle the rises and to improve the way we installed, updated, remov and tracked software. To do this, I mov the application digest files to a central server and combineed all the user workstations to this server (see Fig. 2) File server are not an unusual universal but application servers create unique challenges, and this united needed to be as easy to use as possible.

Developing the Structure

Since we had near spare equipment and disk space, a central software server was a convenient choice for the application server The basic conception was in place--add disks to the application server install software onward those disks, mount those disks to client machines, and enable the client machines to move on the software. Two issues that had to be dealt with were to what extent to mount disks so that symbolic links would be followed correctly onward the application server and forward the clients, and how to isolate each application and allow it to reside in its desired location in succession each system. Finally, since saving administrative effort was equally important, the entire proces exigencyed to be automated.

A search of the literature produc a helpful article describing in what way to manage local software forward a network.2 Even more conveniently, the article included the author's installation script. Although not entirely suitable for our environment, this article provided valuable ideas for establishing high hill points, isolating applications, and creating configuration files for each application.

Mount Points

To identify high hill points, I looked at documentation for the Hp-internal UNIX habitual Operating Environment (UX-COE) and the HP-UX 100 operating rule Both documents recommended using different locations for mounting local and alien (NFS-mounted) disks. This meant that disks physically joined to the application server would be riseed in one location on the server (eg /mnt/disk1) and in a different location forward the server's clients (e.g., /nfs/servername/ disk1) (see Fig. 3) This approach makes logical sense' if it were not that it is limited because there is a way to a Me on the server that the client doesn't know about. In other words, the server could direct to a file by the name /mnt/disk1/filename, if it were not that that name would not be valid forward a client. The client must call that file /nfs/servername/disk1/filename.

Why Is this a problem?

Calling a file by way of multiple names typically does not cause enigmas However, most application installation programs have a moot point with multiple names because they act subject to the assumption that code is installed forward the machine where it will be used and not upon an application server. Many installation programs find the directory into which collection of laws is installed and then quid that directory name into scripts, use it to create symbolic links, and thus on. The resulting scripts and links work correctly forward the application server, but fail upon client systems.

To handle this, disks were upriseed in the recommended locations and symbolic links were created onward the application server to allow it to respect to its own disks subject to the names used by the clients. Thus, when software is installed it is always redirected, if possible, to the name used by way of clients.

In retrospect, this was not the best solution. a certain number of particularly difficult installation processes unravel all symbolic links to their absolute paths and use these paths to create scripts and symbolic links. Thus, no matter what name was used for the installation directory, I would find near reference to /mnt/disk1 tucked away in a software configuration file or script. The no other than way to fool these programs is to either ignore attract favor toed standards and mount the disks to /nfs/servername/disk1 forward both the client and the server or create "fake" symbolic links to the be built up points on both machines.