{
  "schemaVersion": 1,
  "id": "atlas-lesson-topology-job-admission",
  "title": "Job admission, topology and resource fragmentation",
  "category": "learning",
  "format": "Atlas lesson",
  "description": "Places complete jobs onto compatible resources while balancing utilization and service goals.",
  "canonicalUrl": "https://semiaifoundry.com/learn/atlas/topology-job-admission/",
  "publishedDate": null,
  "revisionDate": "2026-10-07",
  "author": null,
  "topic": "Data-center operations",
  "series": null,
  "license": null,
  "rightsUrl": "https://semiaifoundry.com/ai-atlas/licenses.html",
  "version": "1.3.0",
  "status": "Published educational lesson",
  "evidenceScope": "No real user queue or reservation is inferred. Slurm’s named gang-scheduling implementation is not conflated with generic simultaneous parallel-job admission.",
  "contentCoverage": "published-lesson",
  "artifactUrl": "https://semiaifoundry.com/ai-atlas/lesson-data/topology-job-admission-67b7c58b3118f3ac.json",
  "interactiveUrl": "https://semiaifoundry.com/ai-atlas/#topology-job-admission",
  "related": [],
  "supersedes": [],
  "supersededBy": [],
  "contentHash": "4c44096731951d9abac4a3cef5f45d2e7cccb18edd9534f096214dbe12f8418c",
  "contentLength": 4648,
  "jsonUrl": "https://semiaifoundry.com/agents/publications/atlas-lesson-topology-job-admission.json",
  "markdownUrl": "https://semiaifoundry.com/agents/publications/atlas-lesson-topology-job-admission.md",
  "sections": [
    {
      "id": "section-1",
      "title": "Job admission, topology and resource fragmentation",
      "start": 39,
      "end": 425
    },
    {
      "id": "section-2",
      "title": "The mechanism",
      "start": 425,
      "end": 1614
    },
    {
      "id": "section-3",
      "title": "Science and assumptions",
      "start": 1614,
      "end": 1642
    },
    {
      "id": "section-4",
      "title": "Allocation feasibility",
      "start": 1642,
      "end": 2069
    },
    {
      "id": "section-5",
      "title": "Engineering example",
      "start": 2069,
      "end": 2284
    },
    {
      "id": "section-6",
      "title": "Distinctions and tradeoffs",
      "start": 2284,
      "end": 2703
    },
    {
      "id": "section-7",
      "title": "Specifications and interfaces",
      "start": 2703,
      "end": 2988
    },
    {
      "id": "section-8",
      "title": "Interfaces",
      "start": 2988,
      "end": 3171
    },
    {
      "id": "section-9",
      "title": "Check your understanding",
      "start": 3171,
      "end": 3200
    },
    {
      "id": "section-10",
      "title": "Why can an eight-device job wait while eight devices are free?",
      "start": 3200,
      "end": 3635
    },
    {
      "id": "section-11",
      "title": "Evidence and scope",
      "start": 3635,
      "end": 3986
    },
    {
      "id": "section-12",
      "title": "Sources",
      "start": 3986,
      "end": 4268
    },
    {
      "id": "section-13",
      "title": "Continue exploring",
      "start": 4268,
      "end": 4291
    },
    {
      "id": "section-14",
      "title": "Foundations",
      "start": 4291,
      "end": 4511
    },
    {
      "id": "section-15",
      "title": "Related lessons",
      "start": 4511,
      "end": 4648
    }
  ],
  "citations": [
    {
      "title": "Explore in Atlas",
      "url": "https://semiaifoundry.com/ai-atlas/#topology-job-admission"
    },
    {
      "title": "Slurm Topology Guide",
      "url": "https://slurm.schedmd.com/topology.html"
    },
    {
      "title": "Published lesson data",
      "url": "https://semiaifoundry.com/ai-atlas/lesson-data/topology-job-admission-67b7c58b3118f3ac.json"
    },
    {
      "title": "Atlas releases and licensing",
      "url": "https://semiaifoundry.com/ai-atlas/licenses.html"
    },
    {
      "title": "Cluster scheduler and resource allocation",
      "url": "https://semiaifoundry.com/learn/atlas/scheduler/"
    },
    {
      "title": "Data, tensor, pipeline, and state parallelism",
      "url": "https://semiaifoundry.com/learn/atlas/parallelism/"
    },
    {
      "title": "Scale-up, scale-out and management fabrics",
      "url": "https://semiaifoundry.com/learn/atlas/scaleup-scaleout-boundaries/"
    }
  ],
  "recordHash": "a2c465343445cc34228460a03a179e8bb34d85a0968f94efe4fc5a7168a4cea7",
  "contentFormat": "text/markdown",
  "content": "Data-center operations · Atlas v1.3.0\n\n# Job admission, topology and resource fragmentation\n\nPlaces complete jobs onto compatible resources while balancing utilization and service goals.\n\n[Explore in Atlas](https://semiaifoundry.com/ai-atlas/#topology-job-admission)Job allocation across nodes, storage tiers and recovery time\n\nLearning objectiveExplain resource fragmentation and distinguish policy from physical capacity.\n\n## The mechanism\n\nA distributed job needs a compatible set of devices, enough local memory and the connections required by its partitioning. Having many free devices across a cluster does not guarantee that a sufficiently connected group is available. A scheduler considers resource requests, placement, reservations and policy; Slurm’s topology guide shows how connectivity can inform selection. The broad concept of admitting a parallel job together is distinct from Slurm’s particular time-sliced gang-scheduling feature. Fragmentation, priority and reserved headroom can all leave physically healthy hardware temporarily idle. Useful scheduling therefore balances job completion and service targets rather than maximizing a single occupancy number.\n\n1.  Specify each job’s complete resource and communication requirements.\n2.  Identify healthy free resources with compatible memory and software configurations.\n3.  Evaluate whether the required connected group exists under current reservations.\n4.  Choose admission and placement according to explicit completion-time or latency objectives.\n5.  Measure wait time, fragmentation and useful progress before changing scheduling policy.\n\n## Science and assumptions\n\n### Allocation feasibility\n\nAn allocation is feasible only when the job simultaneously meets its resource-count, memory, compatibility, topology and policy constraints. Each check applies to the same requested allocation.\n\nEach term is a Boolean constraint for the requested allocation.\n\nAssumptions A pedagogical constraint model; real schedulers also handle failures, priorities, time and implementation-specific resources.\n\n## Engineering example\n\nSynthetic cluster has four free devices in each of two slow-linked domains. An eight-device job requiring one local domain waits, while two independent four-device jobs may fit immediately.\n\n## Distinctions and tradeoffs\n\nAn idle device is not always wasted capacity; it may be an unusable fragment or reserved headroom for a defined service target.\n\nTight placement can improve communication while increasing queue wait; backfilling smaller jobs uses gaps but must respect promised reservations.\n\nModel shards that fit individually can still wait because their group cannot be placed on a suitable topology.\n\n## Specifications and interfaces\n\nResource request\n\nDevices, memory, topology and duration\n\nA count alone omits important compatibility constraints.\n\nUtilization versus goodput\n\nBusy resources versus useful accepted work\n\nUnproductive retries or deadline misses can inflate activity.\n\n### Interfaces\n\nQueued jobs and policy Input\n\nRequired resources, priorities and placement constraints\n\nAdmitted allocation Output\n\nA compatible job resource set and execution start\n\n## Check your understanding\n\n### Why can an eight-device job wait while eight devices are free?\n\n1.  The scheduler cannot count beyond seven\n2.  The free devices may not form a compatible required topology\n3.  Device counts automatically guarantee placement\n\nAnswer and reasoning\n\nThe free devices may not form a compatible required topology\n\nMemory, software and connectivity constraints can prevent a collection of free devices from forming a valid allocation.\n\n## Evidence and scope\n\nNo real user queue or reservation is inferred. Slurm’s named gang-scheduling implementation is not conflated with generic simultaneous parallel-job admission.\n\nPropagate a stated facility power budget, PUE, readiness and utilization to illustrative work; detailed conversion, battery, cooling and grid models are not included.\n\n## Sources\n\n-   [Slurm Topology Guide](https://slurm.schedmd.com/topology.html)\n\n[Published lesson data](https://semiaifoundry.com/ai-atlas/lesson-data/topology-job-admission-67b7c58b3118f3ac.json) · [Atlas releases and licensing](https://semiaifoundry.com/ai-atlas/licenses.html)\n\n## Continue exploring\n\n### Foundations\n\n-   [Cluster scheduler and resource allocation](https://semiaifoundry.com/learn/atlas/scheduler/)\n-   [Data, tensor, pipeline, and state parallelism](https://semiaifoundry.com/learn/atlas/parallelism/)\n\n### Related lessons\n\n-   [Scale-up, scale-out and management fabrics](https://semiaifoundry.com/learn/atlas/scaleup-scaleout-boundaries/)"
}