LSF version 3.0 includes many new features requested by LSF users, as well as bug fixes to LSF version 2.2. Most of the new features are enhancements to the LSF Batch system.
This chapter summarizes the new features that are related to LSF end users. The new features that concern LSF administration are described in the LSF Administrator's Guide.
When several hosts can satisfy the resource requirements of a job the hosts are ordered by load. However in certain situations it maybe desirable to override this behaviour to give preference to certain hosts even if they are more heavily loaded. For example, your LSF cluster may consist of a few dedicated batch servers and a few desktop machines. A desktop machine can also run jobs when the console user is not using it interactively. You prefer to run on the dedicated servers and only use the desktop if no dedicated server is available.
The -m option of bsub allows you to specify preference by using '+' after the host name. A special hostname 'others' can be used to refer to other hosts not listed. You can also optionally specify a preference level for each host for more flexible preference.
A batch job can be submitted in interactive mode such that all input and output are through the terminal from which the bsub command is issued. This enables the user to run interactive jobs under the resource sharing control supported by LSF Batch. With this support, all the jobs can be under the effective resource sharing control, scheduling policies, and job accounting of LSF Batch.
An interactive batch job is submitted by specifying the -I option of the bsub command.
Shell mode support is provided for running interactive applications through the RES or through LSF Batch. Shell mode support is required to run interactive shells or applications which redefine the CNTL-C and CNTL-Z keys (for example, jove). The -S option to lsrun(1), ch(1), and lsgrun(1) creates the remote task with shell mode support. The -Is option to bsub(1) provides the same feature for interactive batch jobs.
Some jobs have to be started under particular shells or require certain setup steps to be performed before the actual job is executed. The Job Starter feature allows you to specify an executable which will perform the actual execution of the job, doing any necessary setup beforehand. For batch jobs, the job starter can be configured at the queue level by LSF administrator. For non-batch jobs through the RES it can be specified through an environment variable. This feature is particularly useful for users who run jobs in an Atria ClearCase environment.
In previous versions, resource consumption of jobs are only available when a job finishes. With LSF 3.0, it is now possible to see the resource consumption of jobs while they are still running. Such information is collected periodically by LSF Batch daemons and visible by users using the bjobs command.
The CPULIMIT parameter is now enforced at the job-level in addition to the process level. If a job starts multiple processes periodically which do some work and terminate, the system will accumulate the CPU time of previous processes in the job.
If you are concerned about other jobs be dispatched to the same hosts on which your jobs are running, you can use the resource reservation feature of LSF 3.0. Resources can be reserved as specified by resource requirements. No other job will use the reserved resources even if the job which reserved the resources has not consumed them. The amounts of reserved resources can be viewed by users through LSF commands. You can specify both the amount of reservation as well as a duration.
In previous versions, the job pending reasons printed out by bjobs do not give very specific information about related hosts. With 3.0, the related host names are printed out along with the job pending reasons by the bjobs -p -l command.
The following enhancements have been made to more effectively support parallel jobs running through LSF Batch.
LSF 3.0 provides a uniform checkpointing interface to allow users to either use the default checkpointing mechanism, or to integrate their own check-pointing method with LSF Batch. Checkpointing is now accomplished by two external executables: echkpnt and erestart.
Users or LSF administrator can modify or rewrite these executables to interface their own checkpointing facility with LSF Batch.
LSF now recognize a new keyword "hname" in a resource requirement string. This allows more flexibility in host selections. For example, you can use "hname != hostA" to exclude the selection of hostA.
xlsbatch is now customizable. You can customize the menu items of xlsbatch by specifying "customizedMenu" resource in the X resource file. With this feature, you can add/remove/modify menu items in the pull-down menus. Each menu item can be associated with an external executable. So you can define your own actions by changing the menu items.
The -r option has been added to the bjobs command to display jobs in RUN status.
Copyright © 1994-1997 Platform Computing Corporation.
All rights reserved.