The foundation of the site Solaris cluster requires a file server and a Network Information Server (NIS server). Once these are established other machines can be migrated one by one.
Solaris installed on ultrasparc. Configured as a file server. Configured as NIS primary server. No user logins will be permitted to avoid abuse as a computing resource because although it can compute rapidly this would rapidly degrade performance of all other sparcs.
Configure the primary Disk resource which is the RAID array. The array is configured to allow access to all files to continue seamlessly in the event of 3 disk failures in a set of NN.
Solaris installed on the library sparc (sparc5). Configured as a NIS secondary server. This will be the default compute server until lpss1 (a sparc10) is later freed.
Solaris installed on a sparc IPX. Configured as a (not very powerful) compute server.
Aplications are loaded onto the ultrasparc file server begining with a copy taken from the SLO.
Astonomers invited to test (using either the sparc5 or the IPX) applications such as iraf for correct installation.
Subject to the use of the sparc5 as a data reduction tool having been proved to be satisfactory its usage is announced with guest accounts available. User accounts are added on request although such users continue to retain access to the sunOS cluster.
The solaris IPX is presented alongside the sunos IPX machine lpss7 in the INT as an alternative facility for data reduction. After two weeks the sunos IPX is withdrawn. The DAS system at INT does not use lpss7.
The withdrawn IPX is reconstituted as a solaris machine and preparations made to enable it to substitute the JKT sunOS sparc lpss6 which may at that time still be in use as an operational system in the jkt (ie the sparc to which data is transferred from the Perkin-Elmer). On an appropriate s/d night the sunos sparc lpss6 is replaced by the solaris sparc. If this is successful provision of the sunos sparc as a fall back is discontinued after four weeks.
For the first time a significant amount of disk resource is available (approx 36Gb). To guard operations against a single-point failure (the loss of the ultrasparc file server) the IPXs use these disk resources to act as secondary service of applications normally held on the ultrasparc file server. These machines are slow and not ideal for this purpose but would nevertheless enable observing to continue even in the event of a catastrophic failure of the ultrasparc file server. A better machine (lpss1 a sparc10) will eventually take a role but coverage against ultrasparc failure is a priority and cannot wait for lpss1 to be available.
In a day-time exercise the ultrasparc is shutdown and an attempt to use the INT data reduction IPX and the JKT observation system IPX (if not by then an autonomous DAS system) is attempted and relative performance is recorded for future reference.
At this point the only remaining sparcs on site running sunOS are lpss1 and lpss2 and there is no night-time operational dependence on these machines. NIS and DNS are reconfigured to define ing.iac.es to be the ultrasparc and simultaneously mail aliases are defined such that mail is delivered to lpss1 for only those users who have not yet acquired an account on the Solaris cluster.
Users accounts are relocated to Solaris as fast as installation of any missing applications which users need permits.
After all users have been relocated to the solaris cluster then simultaneously lpss1 and lpss2 are withdrawn from service.
Finally lpss1 which is a reasonably fast machine is free and takes on the role of secondary file server in conjunction with between one and three IPXs. The library sparc5 is deliberately excluded from any role in secondary file serving as it must at all times be free to function as a spare for any of the sun4m architecture machines in mission-critical roles.