Choosing a QOS for Your Job¶
Use this page to help decide which Quality of Service (QOS) to request for your Slurm job on Perlmutter.
flowchart TD
A([What are you trying to do?]) --> B[Submit a compute job]
A --> C[Need guaranteed fast<br/>turnaround or have a<br/>realtime allocation]
A --> D[Move data between<br/>filesystems or systems]
A --> E[Run a recurring job<br/>on a schedule via scrontab]
B --> F["See standard QOS chart"]
C --> I["See Premium & Realtime chart"]
D --> G[xfer]
E --> H[cron]
click F "#standard-qoses" "Jump to Standard QOS chart"
click I "#premium-and-realtime" "Jump to Premium & Realtime chart"
click G "../examples#xfer-qos" "xfer QOS docs"
click H "../workflow/scrontab" "cron QOS docs"
classDef start fill:#002D56,stroke:#0B2139,color:#EABB5B,font-weight:bold;
classDef question fill:#336FAD,stroke:#0B2139,color:#F0F3F3,font-weight:bold;
classDef qos fill:#BACC4F,stroke:#0B2139,color:#002D56,font-weight:bold;
classDef link fill:#64666A,stroke:#0B2139,color:#F0F3F3,font-weight:bold,font-style:italic;
class A start;
class B,C,D,E question;
class G,H qos;
class F,I link; Standard QOSes¶
flowchart TD
A([Start: Submitting a<br/>compute job]) --> B[Do you need an<br/>interactive session?]
B -- Yes --> C[Need less than a full node?<br/>e.g. 1-2 GPUs, single-threaded CPU work]
C -- Yes --> D[shared_interactive]
C -- No --> E[interactive]
B -- No --> F[Quick test/debug run?<br/>≤30 min, small node count]
F -- Yes --> G[debug]
F -- No --> H[Need less than a full node?<br/>e.g. 1-2 GPUs, single-threaded CPU work]
H -- Yes --> I[shared]
H -- No --> J[Can the work tolerate<br/>preemption/requeue?]
J -- Yes --> K[preempt]
J -- No --> L[regular]
click E "../../jobs/interactive" "interactive QOS docs"
click G "../../policies/resource-usage#debug" "debug QOS docs"
click K "../../policies/resource-usage#preempt" "preempt QOS docs"
classDef start fill:#002D56,stroke:#0B2139,color:#EABB5B,font-weight:bold;
classDef question fill:#336FAD,stroke:#0B2139,color:#F0F3F3,font-weight:bold;
classDef qos fill:#BACC4F,stroke:#0B2139,color:#002D56,font-weight:bold;
class A start;
class B,C,F,H,J question;
class D,E,G,I,K,L qos; Premium and Realtime¶
flowchart TD
A([Start: Need faster than<br/>regular turnaround?]) --> B[Do you have an approved<br/>realtime allocation?<br/>e.g., facility/experiment support]
B -- Yes --> C[realtime]
B -- No --> D[Are you willing to accept a<br/>2x charge factor for<br/>higher scheduling priority?<br/>Has your PI enabled<br/>premium QOS use for you?]
D -- Yes --> E[premium]
D -- No --> F["See standard QOS chart,<br/>use regular or preempt"]
click C "../../policies/resource-usage#realtime" "realtime QOS docs"
click E "../../policies/resource-usage#premium" "premium QOS docs"
click F "#standard-qoses" "Jump to Standard QOS chart"
classDef start fill:#002D56,stroke:#0B2139,color:#EABB5B,font-weight:bold;
classDef question fill:#336FAD,stroke:#0B2139,color:#F0F3F3,font-weight:bold;
classDef qos fill:#BACC4F,stroke:#0B2139,color:#002D56,font-weight:bold;
classDef link fill:#64666A,stroke:#0B2139,color:#F0F3F3,font-weight:bold,font-style:italic;
class A start;
class B,D question;
class C,E qos;
class F link; QOS Summary¶
| QOS | Use Case |
|---|---|
debug | Short test jobs |
interactive | Interactive salloc sessions |
regular | Standard production workloads |
premium | Faster turnaround at a higher charge |
preempt | Work amenable to premption (via checkpoint/restart or otherwise) |
realtime | Jobs with an external real-time component |
xfer | Data transfer |
Still need help?¶
Please see our QOSes and Charges page for more information about the details of each QOS. If your workload does not fit within the constraints of the available QOSes, you may submit a reservation request to get a resource allocation that matches your needs. You can also submit a ticket at our help desk with any questions about submitting jobs.