Why does a job scheduler matter on a shared GPU cluster?
Why B is the answer
It allocates GPUs fairly and queues the work
Schedulers share limited resources.
In detail
- Schedulers share limited resources fairly.
- It stops one team taking everything.
Use case
An illustrative scenario, not a real incident.
Ten teams share 20 GPUs, and a scheduler queues work and shares them fairly.
See it in different ways
flowchart TD
A{"What does the workload look like?"} -->|"Batch jobs"| B["Slurm"]
A -->|"Containers and services"| C["Kubernetes"]mindmap
root((Operations))
Monitoring
nvidia-smi
DCGM
Scheduling
Slurm
Kubernetes
Provisioning
Base Command Manager
Sharing
MIG
vGPUflowchart LR S0["Jobs submitted"] --> S1["Queued"] --> S2["GPUs allocated"]
Key point: Schedulers share limited GPUs.
Why the other options are wrong
- A. It writes the application code for each user
- They do not write code.
- C. It speeds up each GPU's clock when it runs
- They do not change clocks.
- D. It compresses the datasets on shared storage
- They do not compress data.