How to choose a partition in O2
- Kathleen Chappell
- Gaby Rodriguez
- Neil Coplan
In O2 we minimized the number of available partitions (or queues) to simplify the job submission process.
You can use the flow chart below to determine which partition is best for the jobs you plan to run.
Note 1: all partitions can currently be used to request interactive sessions; the interactive partition has a dedicated set of nodes and higher priority.
Note 2: all partitions can run single or multi-core jobs. For more information about parallel jobs in O2 read our dedicated wiki page
Details about the available partitions are reported in the table below
Partition | Job Type | Priority | Max cores per job | Max runtime limit | Min runtime limit | Notes |
|---|---|---|---|---|---|---|
interactive | interactive | 14 | 20 | 12 hours | n/a | 2 job limit, 20 core/job limit, 250GiB / job memory limit, default memory 4GB |
short | batch & interactive | 12 | 20 | 12 hours | n/a | 20 core/job limit, 250GiB / job memory limit |
medium | 6 | 20 | 5 days | 12 hours | 20 core/job limit, 250GiB / job memory limit | |
long | 4 | 20 | 30 days | 5 days | 20 core/job limit, 250GiB / job memory limit | |
mpi | 12 | 640 | 5 days | n/a | invite-only. Email rchelp | |
priority | 14 | 20 | 30 days | n/a | limit 2 jobs running at once, 20 core per job limit 250GiB / job memory limit | |
transfer | n/a | 4 | 5 days | n/a | limit of a 5 concurrently cores per user for transfers between O2 and See File Transfer for more information. Invite-only. Email rchelp. | |
gpu, gpu_quad | n/a | 20 | 5 days | n/a | additional limits apply to the GPU partition, see the Using O2 GPU resources page for more details | |
gpu_mpi_quad | n/a | 240 | 5 days | n/a | invite-only. Email rchelp | |
gpu_requeue | n/a | 20 | 5 days | n/a | see O2 GPU Re-Queue Partition for additional details | |
highmem | n/a | 16 | 5 days | n/a | invite-only. Email rchelp |
Although Slurm interprets units as KiB, MiB, and GiB in the background, O2 users must use K, M, or G as valid units for the memory (--mem) argument. KB, MB, or GB are other alternative units.
Flowchart Description
Decision Pathways:
Does your job need access to
/files?Yes: Use the dedicated transfer cluster (
transfer.rc.hms.harvard.edu) for copying/moving data.No: Proceed to the next question.
Does your job need more than 200GiB of memory?
Yes: Use the highmem partition.
No: Proceed to the next question.
Does your job need access to GPU resources?
Yes: Choose from:
gpu
gpu_quad
gpu_requeue
gpu_mpi_quad
No: Proceed to the next question.
Are your jobs MPI-jobs or capable of running on distributed memory systems?
Yes: Use the mpi partition.
No: Proceed to the next question.
Are you trying to run 1 or 2 interactive jobs?
Yes: Use the interactive partition.
No: Proceed to the next question.
Do you need to run only 1 or 2 high-priority jobs?
Yes: Use the priority partition.
No: Proceed to the next question.
Are your jobs going to run for less than 12 hours?
Yes: Use the short partition.
No: Proceed to the next question.
Are your jobs going to run for less than 5 days?
Yes: Use the medium partition.
No: Use the long partition.
Partitions and Their Characteristics:
“interactive”
Job Type: Interactive
Priority: 14
Max Cores per Job: 20
Max Runtime: 12 hours
Notes: 2 job limit, 20 cores per job, 250GiB memory limit, default memory 4GB
“short”
Priority: 12
Max Cores per Job: 20
Max Runtime: 12 hours
Notes: 20 core/job limit, 250GiB memory limit
“medium”
Priority: 6
Max Cores per Job: 20
Max Runtime: 5 days
Min Runtime: 12 hours
Notes: 20 core/job limit, 250GiB memory limit
“long”
Priority: 4
Max Cores per Job: 20
Max Runtime: 30 days
Min Runtime: 5 days
Notes: 20 core/job limit, 250GiB memory limit
“mpi”
Job Type: Batch
Priority: 12
Max Cores per Job: 640
Max Runtime: 5 days
Notes: Invite-only, email rchelp
“priority”
Job Type: Batch & Interactive
Priority: 14
Max Cores per Job: 20
Max Runtime: 30 days
Notes: Limit 2 jobs running at once, 20 core/job limit, 250GiB memory limit
“transfer”
Priority: n/a
Max Cores per Job: 4
Max Runtime: 5 days
Notes: Limit of 5 concurrent cores per user, for file transfers between O2 and
/n/files
“gpu” and “gpu_quad”
Priority: n/a
Max Cores per Job: 20
Max Runtime: 5 days
Notes: Additional limits apply, see the Using O2 GPU resources page
“gpu_mpi_quad”
Priority: n/a
Max Cores per Job: 240
Max Runtime: 5 days
Notes: Invite-only, email rchelp
“gpu_requeue”
Priority: n/a
Max Cores per Job: 20
Max Runtime: 5 days
Notes: See O2 GPU Re-Queue Partition for details
“highmem”
Priority: n/a
Max Cores per Job: 16
Max Runtime: 5 days
Notes: Invite-only, email rchelp
Important
The interactive and short partitions allow jobs up to 12 hours, while medium, long, and mpi allow jobs up to 5–30 days.
The mpi partition supports 640 cores per job, making it suitable for large-scale parallel jobs.
The priority partition allows long jobs but has a 2-job concurrent limit.
Transfer partition is dedicated to data transfer operations, with a limit of 5 concurrent cores per user.
GPU partitions have additional constraints and require checking separate documentation.
Some partitions, like mpi, gpu_mpi_quad, and highmem, are invite-only.
There is a limit on the total CPU-hours that can be reserved by a single lab at any given time. This limit was introduced to prevent a single lab from locking down a large portion of the cluster for extended periods of time. This limit will become active only if multiple users in the same lab are allocating a large portion of the O2 cluster resources. This can for example happen if few users have thousands of multi-day or hundreds of multi-week running jobs. When this limit becomes active the remaining pending jobs will display the message AssocGrpCPURunMinute.
Note: The "gpu" partition has additional limits that might trigger the above message, for more details about the "gpu" partition please refer to the "Using O2 GPU resources" wiki page