Queues

 

Home
Clusters
Queues
Jobs
Commands
Manuals
FAQ

All LSF jobs run in queues. Each queue can have several attributes, including:

the name, which uniquely identifies the queue
a list of the hosts which can run jobs from the queue
limits on the number of hosts, jobs, users, processors, etc.
standard UNIX limits such as memory, CPU, processes, etc.
the scheduling policy used, such as first-come-first-served, fairshare, and exclusive
load-sharing thresholds
the UNIX nice value, which determines the relative priority of the processes in the job

A description of the queues available in the ATN clusters can be found below. You can also use the bqueues command to list the properties of a specific queue.  For example,  you could type "bqueues -l batch" to find out more about the batch queue.

batch

This is a general-purpose queue for running non-interactive jobs in the background.  It is open to all users and can dispatch jobs to most of the server hosts in the cluster. Each job in the batch queue can use up to 10 gigabytes of memory (if available).

If you do not specify a specific queue your job will go to the batch queue by default.

express

This queue is designed for non-interactive jobs that need no more than five minutes of CPU time or 10 minutes of "wall clock" time. If a job in this queue is still running after that, it is terminated.

The express queue is not available on all hosts.

interactive

This is a general-purpose queue for running interactive jobs. Like the batch queue, the interactive queue is open to all users and can dispatch jobs to most of the server hosts  in the cluster. Each user can run up to four jobs at once in the interactive queue.  Each job in the interactive queue can use up to 10 gigabytes of memory (if available).

Jobs in the interactive queue run at a higher priority relative to the other queues in order to improve response time.

parallel

This queue is intended for applications that have been designed to run in parallel on multiple CPUs rather than serially on a single CPU. These CPUs may be on the same host or on different hosts. Each parallel job can use up to 10 gigabytes of memory (if available) .

The parallel queue is open to all users but will dispatch jobs only to selected hosts. 

Jobs in the parallel queue can be run either in the background or interactively in the foreground.

 

Last updated Tuesday, April 02, 2002 03:18 PM