{
  "slug": "memory-stranding",
  "term": "Memory stranding",
  "aliases": [],
  "advantages": [],
  "limitations": [],
  "supersedes": [],
  "supersededBy": [],
  "firstSeen": null,
  "lastSeen": null,
  "updated": "2026-11-16",
  "definition": "Memory stranding is available memory capacity that cannot be allocated to an otherwise relevant workload because another placement resource or constraint is exhausted.",
  "note": "In Pond's motivating case, a server has unallocated DRAM after its CPU cores have been allocated. This differs from untouched memory inside an already provisioned VM: one concerns placement capacity, the other concerns access to assigned capacity. [paper]\n\nConstructed example: host A has 0 free cores and 64 GiB local memory; host B has 8 free cores and 16 GiB local memory. A new VM needs 4 cores and 32 GiB. No host can admit it without pooling, even though 80 GiB is free in total. In a second configuration, moving 16 GiB of A's unallocated capacity to an accessible pool lets B use 16 GiB local plus 16 GiB pooled memory. Total free capacity remains 80 GiB; it has not been created by pooling. [examples]\n\nLocal-memory constraint: if that VM requires at least 24 GiB local memory, B cannot admit it even with the 16 GiB pool. The minimum-local requirement is an input assumption, not a measured latency guarantee. The executable predicate tests one VM against a static snapshot, with integer GiB capacity and no oversubscription. [examples] [code]\n\nEngineering implication: record which resources are free, which host can access them and which workload constraints apply. Summing free bytes across hosts is insufficient evidence of allocatable capacity; neither an idle-page count nor a pool's name supplies the missing placement contract.",
  "status": "shipping",
  "related": [
    "pond",
    "cxl-memory",
    "cxl-memory-pooling"
  ],
  "citations": [
    {
      "id": "paper",
      "title": "Li et al. Pond: CXL-Based Memory Pooling Systems for Cloud Platforms. ASPLOS 2023, pp. 574-587. DOI 10.1145/3575693.3578835; sections 2-6.",
      "url": "https://www.microsoft.com/en-us/research/wp-content/uploads/2022/10/2023_Pond_asplos23_official_asplos_version.pdf",
      "accessed": "2026-11-16"
    },
    {
      "id": "examples",
      "title": "Hesela: constructed single-VM admission examples and latency-ratio arithmetic (JSON)",
      "url": "https://hesela.dev/research/pond-admission.json",
      "accessed": "2026-11-16"
    },
    {
      "id": "code",
      "title": "Hesela: executable admission predicate and latency arithmetic (JavaScript)",
      "url": "https://hesela.dev/research/pond-admission.mjs",
      "accessed": "2026-11-16"
    }
  ],
  "provenance": [
    "The definition follows the resource-stranding distinction in Pond; the capacity examples are explicitly constructed by Hesela, not observations of Azure or a hardware experiment.",
    "Shipping here classifies an established systems concept; it does not claim that Pond is a shipping product."
  ]
}
