Preparing the system to install IBM Spectrum LSF RTM

IBM Spectrum LSF RTM (RTM) can be installed on various operating systems. This guide contains all information necessary for installing RTM. Some sections are optional. Refer only to those topics that apply to your selected installation path.

All installation paths involve the following main steps:
  1. Preparing to install IBM Spectrum LSF RTM
  2. Installing IBM Spectrum LSF RTM
  3. Activating IBM Spectrum LSF RTM

System Sizing

The most significant factors when sizing your system for RTM are:
  • The number of jobs per day across all clusters
  • The number of hosts to be monitored
  • The number of clusters to be monitored
  • The polling frequencies for each cluster
  • The number of concurrent logins to be supported
  • The RRD file data retention wanted (for example, 1 minute rows, 5 minute rows, 30 minute rows, 2 hour rows, and 1 day rows)
  • The amount of historical job data to be retained

Database

Consider the following points when choosing and preparing your database for RTM:
  • Database sizing depends on number of jobs per day and job detail data retention.
  • Plan for at least 4 KB per job per host.
  • Consider the length of running jobs as a factor on job storage requirements.
  • It is not recommended to use the NFS protocol for database since the files are accessed frequently.

Cacti RRDfiles

Consider the following points when planning your RTM installation:
  • Cacti RRD file size depends on data retention policies. The default data retention results in smaller RRD files. Therefore, assume 1-2 MB per host and 30 MB or more per cluster, depending on the number of users, projects, or queues.
  • RRD files can be very large, depending on your data retention requirements. Refer to https://www.rrdtool.org and https://docs.cacti.net/ for documentation on setting up and configuring data retention using RRDtool.
  • RRD file size must be able to fit in disk cache. For example, if you are expecting 10 GB of RRD files, then you must have a system with more than 20 GB of memory. Try to allocate more memory for better results.
  • It is not recommended to use the NFS protocol for RRD files since these files are accessed frequently.

Job Graphs

Consider the following points when planning your RTM installation:
  • Job Graphs, in general, are created on demand. Therefore, data storage requirements vary depending on the number of users that are viewing job graphs.
  • Each set of job graphs results in 2 MB of file storage. However, if job graph archiving is enabled and you have high volume clusters, then it can result in terabytes of RRD files. For example, 10,000 jobs per day results in a requirement of 20 GB of storage per day for archiving.

Syslog Volume

The syslog database can become large depending on your per host log level, the number of hosts, and your discard rules. Currently, there is no recommended estimate for the syslog database sizing.

Installing RTM with other LSF family products

If you are planning to install RTM on a host that is running IBM Spectrum LSF Application Center or an LSF Suite product (LSF Suite for Enterprise, LSF Suite for HPC, or LSF Suite for Workgroups) as a Gui_role machine, note the following:

  • The RTM installation optimizes the mysql/mariadb setting which requires removal of the InnoDB logs. If you have IBM Spectrum LSF Application Center or an LSF Suite product installed on that host already, you can either:
    • Set the parameter OPTIMIZE_MYSQL=N to avoid the removal OR
    • Back up the IBM Spectrum LSF Application Center database (for example, using mysqldump) and shut down mysql before running the RTM installer. Then remove the new InnoDB logs. After RTM is installed, then restore the IBM Spectrum LSF Application Center database.

Other considerations

Please consider the following when planning your RTM installation:

  • Do not install RTM, its database or any part of its components on NFS. It is not advised due to variability of service levels and the shared nature of the service.
  • Spectrum Scale is not advised due to variability of service levels and the shared nature of the service.
  • VMs are not advised due to variability of service levels and the shared nature of the service.