Skip to content

SLURM Partitions

Partition are a logical set of nodes with a specific technology. For our purposes in AWS, this most logically equates to a particular instance EC2 instance type & size.

Jobs are computational work units which specify resources (cores, memory, time limit, …) and are submitted to specific partition.

Job steps are one or more sets of (possibly parallel) tasks within a job.

The partitions can be considered job queues, each of which has an assortment of constraints such as job size limit, job time limit, users permitted to use it, etc. Priority-ordered jobs are allocated nodes within a partition until the resources (nodes, processors, memory, etc.) within that partition are exhausted.

Partition Configurations

The following table illustrates the default sizes and shapes of partitions that are created for each research team during the onboarding process. It is relatively easy for us to add or remove partitions, though there are limitations on the total number of partitions we can have on one cluster.Adjustments to an existing partition can be done during the maintenance periods only.

Note: This table is subject to change and may become out of date quickly. Use the Checking Partition Options instructions below to get the latest information.

Partition Instance Family Characteristics vCPUs totRAM Instance Storage (GB) Network Bandwidth (Gbps) EBS Bandwidth (Gbps) nStandby Cost ($/hr)
serial c6i.large compute optimized
Intel processors		 2 4 n ≤ 12.5 ≤ 10 0 $0.085
small c6i.2xlarge compute optimized
Intel processors		 8 16 n ≤ 12.5 ≤ 10 0 $0.340
dorieh, med* c6i.4xlarge compute optimized
Intel processors		 16 32 n ≤ 12.5 ≤ 10 1 $0.680
med* c6i.8xlarge compute optimized
Intel processors		 32 64 n 12.5 10 1 $1.360
large x2iezn.2xlarge memory instensive
Intel processors; extra storage/memory; high CPU frequency; network/EBS optimized		 8 256 n ≤ 25 3.17 0 $1.668
xl x2iezn.4xlarge memory instensive
Intel processors; extra storage/memory; high CPU frequency; network/EBS optimized		 16 512 n ≤ 25 4.75 0 $3.336
ultra x2iedn.16xlarge memory instensive
Intel processors; extra storage/memory; instance storage; network/EBS optimized		 64 2048 1 x 1900 NVMe SSD 50 40 0 $13.338
gpu p3.2xlarge gpu accelerated
l40s g6e.4xlarge 1 x NVIDIA L40S GPU/48 GB RAM 16 128 1 x 600 NVMe SSD 20 8 0 $3.004
a100 p4d.24xlarge 8 x NVIDIA A100 GPU/40 GB RAM 96 1152 8 x 1000 NVMe SSD Std Network + Fiber Adapter 19 0 $21.958

Note: 1. vCPUs = a hyperthread. 2 vCPUs = 1 CPU core. 2. Costs as noted here are for on-demand instances. We can lower costs by either specifying a reserve. contract or earmarking certain partitions as spot instances (~= serial requeue).
3. 'med' queues could have 4xlarge or 8 xlarge sizes. Currently only the urcdtest-med partition is this lower size.

Checking Partition Options

There are no general purpose (cross-cluster) partitions; only group-specific ones. By default you will see only the partitions created for your research group.

sinfo - reports the state of partitions and nodes managed by Slurm. It has a wide variety of filtering, sorting, and formatting options.

[jharvard@ip-10-3-2-176 ~]$ sinfo
PARTITION       AVAIL  TIMELIMIT  NODES  STATE NODELIST
urcdtest-serial    up   infinite     20  idle~ urcdtest-serial-dy-urcdtest-serial-cr-0-[1-20]
urcdtest-med       up   infinite      4  idle~ urcdtest-med-dy-urcdtest-med-cr-0-[1-4]
urcdtest-med       up   infinite      1   idle urcdtest-med-st-urcdtest-med-cr-0-1

Note: idle~ represents nodes that are not spun up (not provisioned), while idle represents nodes that are spun up (are provisioned). Provisioning of nodes typically takes 2-5 minutes, depending upon its size.

To see details about partitions, use the scontrol command as follows:

[jharvard@ip-10-3-2-176 ~]$ scontrol show partition
 PartitionName=urcdtest-serial
    AllowGroups=urc-urcdtest-lab AllowAccounts=ALL AllowQos=ALL
    AllocNodes=ALL Default=NO QoS=N/A
    DefaultTime=NONE DisableRootJobs=NO ExclusiveUser=NO GraceTime=0 Hidden=NO
    MaxNodes=UNLIMITED MaxTime=UNLIMITED MinNodes=0 LLN=NO MaxCPUsPerNode=UNLIMITED
MaxCPUsPerSocket=UNLIMITED
    NodeSets=urcdtest-serial_nodes
    Nodes=urcdtest-serial-dy-urcdtest-serial-cr-0-[1-20]
    PriorityJobFactor=1 PriorityTier=1 RootOnly=NO ReqResv=NO OverSubscribe=NO
    OverTimeLimit=NONE PreemptMode=OFF
    State=UP TotalCPUs=40 TotalNodes=20 SelectTypeParameters=NONE
    JobDefaults=(null)
    DefMemPerNode=UNLIMITED MaxMemPerNode=UNLIMITED
    TRES=cpu=40,mem=77820M,node=20,billing=40
    ResumeTimeout=GLOBAL SuspendTimeout=GLOBAL SuspendTime=GLOBAL PowerDownOnIdle=NO

There are a few parameters that you might consider requesting an update from the default:

  • MaxTime=UNLIMITED, which sets at the parition level the max time for a job. Otherwise, this can be set per job in a user's job script.
  • MaxNodes=UNLIMITED, which sets the maximum number of nodes per user. If you don't want a single user to dominate all available resources, this can be reduced to a fraction of the TotalNodes.

Checking Node Options

To see details about a particular node from your sinfo nodelist, use the scontrol command as follows:

[jharvard@ip-10-3-2-176 ~]$ scontrol show node urcdtest-serial-dy-urcdtest-serial-cr-0-1
 NodeName=urcdtest-serial-dy-urcdtest-serial-cr-0-1 Arch=x86_64 CoresPerSocket=1 
    CPUAlloc=0 CPUEfctv=2 CPUTot=2 CPULoad=0.12
    AvailableFeatures=dynamic,c6i.large,urcdtest-serial-cr-0
    ActiveFeatures=dynamic,c6i.large,urcdtest-serial-cr-0
    Gres=(null)
    NodeAddr=urcdtest-serial-dy-urcdtest-serial-cr-0-1
NodeHostName=urcdtest-serial-dy-urcdtest-serial-cr-0-1 Version=23.02.6
    OS=Linux 5.10.230-223.885.amzn2.x86_64 #1 SMP Tue Dec 3 14:36:00 UTC 2024 
    RealMemory=3891 AllocMem=0 FreeMem=1937 Sockets=2 Boards=1
    State=IDLE+CLOUD+POWERED_DOWN ThreadsPerCore=1 TmpDisk=0 Weight=1000 Owner=N/A MCS_label=N/A
    Partitions=urcdtest-serial 
    BootTime=2025-01-10T23:20:37 SlurmdStartTime=2025-01-10T23:22:58
    LastBusyTime=Unknown ResumeAfterTime=None
    CfgTRES=cpu=2,mem=3891M,billing=2
    AllocTRES=
    CapWatts=n/a
    CurrentWatts=0 AveWatts=0
    ExtSensorsJoules=n/s ExtSensorsWatts=0 ExtSensorsTemp=n/s

ActiveFeatures provides a clue into the instance type: c6i.large

CfgTRES provides the number of cores and memory: cpu=2,mem=3891M

See the AWS EC2 instance types for the full tech specs of nodes.

For example, c6i.large contains 3.5Ghz Intel Xeon 8375C with 2 cores, 4 GB of memory and 25 Gbps network.

See All Partitions

The list and composition of the partitions changes with each maintenance.

To see the must up-to-date list, simply list them with sinfo -a

[bash@ip-10-3-2-176 ~]$ sinfo -a
PARTITION        AVAIL  TIMELIMIT  NODES  STATE NODELIST
dominici-serial*    up   infinite     19  idle~ dominici-serial-dy-dominici-serial-cr-0-[1-19]
dominici-serial*    up   infinite      1   idle dominici-serial-st-dominici-serial-cr-0-1
dominici-small      up   infinite     16  idle~ dominici-small-dy-dominici-small-cr-0-[1-16]
dominici-med        up   infinite     10  idle~ dominici-med-dy-dominici-med-cr-0-[1-10]
dominici-large      up   infinite      5  idle~ dominici-large-dy-dominici-large-cr-0-[1-5]
dominici-xl         up   infinite      3  idle~ dominici-xl-dy-dominici-xl-cr-0-[1-3]
dominici-ultra      up   infinite      2  idle~ dominici-ultra-dy-dominici-ultra-cr-0-[1-2]
dominici-gpu        up   infinite      2  idle~ dominici-gpu-dy-dominici-gpu-cr-0-[1-2]
schwartz-serial     up   infinite     20  idle~ schwartz-serial-dy-schwartz-serial-cr-0-[1-20]
schwartz-small      up   infinite     16  idle~ schwartz-small-dy-schwartz-small-cr-0-[1-16]
schwartz-med        up   infinite     10  idle~ schwartz-med-dy-schwartz-med-cr-0-[1-10]
schwartz-large      up   infinite      5  idle~ schwartz-large-dy-schwartz-large-cr-0-[1-5]
schwartz-xl         up   infinite      2  idle~ schwartz-xl-dy-schwartz-xl-cr-0-[1,3]
schwartz-xl         up   infinite      1  alloc schwartz-xl-dy-schwartz-xl-cr-0-2
schwartz-ultra      up   infinite      2  idle~ schwartz-ultra-dy-schwartz-ultra-cr-0-[1-2]
schwartz-gpu        up   infinite      2  idle~ schwartz-gpu-dy-schwartz-gpu-cr-0-[1-2]
urcdtest-serial     up   infinite     20  idle~ urcdtest-serial-dy-urcdtest-serial-cr-0-[1-20]
urcdtest-med        up   infinite      4  idle~ urcdtest-med-dy-urcdtest-med-cr-0-[1-4]
urcdtest-med        up   infinite      1   idle urcdtest-med-st-urcdtest-med-cr-0-1
dominici-dorieh     up   infinite      2  idle~ dominici-dorieh-dy-dominici-dorieh-cr-0-[1-2]

Updated Oct 17, 2025