Slurm Flashcards

A coordinated set of networked compute resources managed as one scheduling domain.

A submitted unit of computational work, typically represented as a job or set of jobs.

The decision function that maps queued jobs to available resources over time.

The policy process that resolves contention among competing users.

Primary controller daemon responsible for scheduling and state orchestration.

Node-resident daemon that launches and monitors tasks on compute nodes.

Accounting daemon that persists usage records to a database backend.

Logical layer responsible for coordination and policy decisions.

Trackable resources such as CPU, memory, GPU, and custom generic resources.

Generic resources, commonly used for GPUs or specialized devices.

Binding of tasks to specific CPUs or NUMA regions.

Non-Uniform Memory Access; memory latency depends on placement locality.

Operational status such as idle, allocated, drain, down, or unknown.

A policy-scoped subset of nodes presented as a queue target.

State indicating intentional exclusion from new scheduling.

Rules limiting which users or accounts can submit to a partition.

Primary scheduler and cluster topology configuration file.

Device-level generic resource inventory definitions.

Runtime enforcement policy for CPU, memory, and device isolation.

Deviation between intended and actual deployed settings.

Credential signing service used for authenticated communication in Slurm.

Policy decision about whether an authenticated identity may perform an action.

Security principle granting only the minimum required permissions.

Administrative separation preventing one tenant from affecting another.

Command for cluster and partition state visibility.

Command for queued and running job visibility.

Command for script-based batch submission.

Command for job termination by identifier or filter.

Job admitted to queue but not yet allocated resources.

Job has active resource allocation and executing tasks.

Terminal state with successful execution status.

Terminal state due to policy-enforced runtime limit breach.

Script-defined workload submitted for deferred execution.

Allocation enabling immediate user-driven command execution.

Parametrized set of homogeneous job instances.

Execution subdivision inside an allocated parent job.

First-in-first-out ordering based on submission sequence.

Scheduling strategy filling idle windows without delaying reservations.

Policy model adjusting priority using historical usage.

Composite ranking using weighted scheduling factors.

Administrative coefficient controlling influence of a priority factor.

Temporal reduction of historical usage influence over time.

Scaling factors to comparable ranges before weighted summation.

Usage-based contribution to current job priority.

Policy object encoding limits, priorities, and preemption behavior.

Policy action allowing one workload to displace another based on rules.

Maximum wall-clock runtime allowed under a policy class.

Upper bound on allocatable resources for a user, account, or QoS class.

Dependency requiring predecessor job start before dependent eligibility.

Dependency requiring predecessor successful completion.

Dependency requiring predecessor failure state.

Constraint ensuring one active job of a given key context.

Kernel mechanism for hierarchical resource control and accounting.

Slurm plugin integrating job/task boundaries with cgroup enforcement.

Policy control limiting memory usage within cgroups.

Allowed device access set for constrained workloads.

Generic resource abstraction used for GPUs and specialized devices.

Multi-Instance GPU partitioning for hardware-level sub-device isolation.

Mapping tasks to specific GPU identifiers for locality control.

Placement logic incorporating physical interconnect structure.

Single logical job containing components with different resource vectors.

One resource-homogeneous segment inside a heterogeneous submission.

Dependency relation between computational stages in a pipeline.

Data placement relationship affecting transition overhead between stages.

Daemon responsible for collecting and persisting accounting records.

Command-line interface for querying job-level accounting records.

Command for aggregated utilization and accounting reports.

Attribution of resource consumption cost to organizational entities.

Number of jobs waiting for allocation at a given observation point.

Duration from submission until execution start.

Controller time to evaluate and dispatch eligible jobs.

Fraction of available resources actively consumed by workloads.

Active scheduler authority in normal operation.

Standby controller prepared for failover takeover.

Controlled transfer of scheduler authority to backup infrastructure.

Propagation of controller state required for coherent takeover.

Logical linkage of multiple Slurm clusters for coordinated scheduling semantics.

Policy-aware placement decisions spanning cluster boundaries.

Capability to place workloads where capacity exists across federated members.

Usage visibility and governance across multiple cluster domains.

Transition node to low-power state when not needed for active scheduling.

Transition node back to schedulable state in response to demand.

Ability to vary active capacity based on workload intensity.

Time required to activate capacity and make it schedulable.

Process of creating and initializing cluster infrastructure resources.

Property of automation where repeated execution yields consistent final state.

Detection and remediation of state divergence over time.

Initial configuration sequence required to make nodes operational.

HPC-friendly container runtimes with user-space execution patterns.

Property of fixed container content across executions.

Separation of software libraries to avoid cross-workload conflicts.

Ability to run the same workload in multiple environments with consistent behavior.

NVIDIA communication library optimized for multi-GPU collective operations.

Message Passing Interface standard for distributed process coordination.

NVIDIA parallel computing platform for GPU-accelerated execution.

Model training across multiple processes and often multiple nodes.

Placement strategy respecting network and hardware locality structure.

Proximity relationship affecting latency and bandwidth between compute resources.

Interval in which the scheduler evaluates queue and dispatch opportunities.

Dominant constraint limiting end-to-end workload throughput or latency.