{"resourceId":"tacc-horizon-user-guide-vast-home-scratch-20261110","versions":[{"version":"external-a7a0a9c419a568b6fbe154abce204769f5af8494c6516c560247c441e6896be6","resource":{"id":"tacc-horizon-user-guide-vast-home-scratch-20261110","title":"TACC's Horizon User Guide (updated Nov 10, 2026) puts NSF LCCF early users on shared VAST $HOME and $SCRATCH, with $WORK deferred to late 2026","organization":"Texas Advanced Computing Center (TACC), The University of Texas at Austin","sector":"Public research university / NSF-funded national research computing facility","geography":"Texas, United States (national user base)","publishedAt":"2026-11-10","publicationDate":"2026-11-10","eventDate":"2026-11-10","sourceName":"TACC HPC Documentation","sourceLabel":"Operator user documentation (draft)","sourceUrl":"https://docs.tacc.utexas.edu/hpc/horizon/","evidenceClass":"public-sector-association","outcomeClass":"emerging","topics":["infrastructure","operating-model","accessibility-workforce"],"finding":"The Horizon User Guide, last updated November 10, 2026 and labeled a draft in progress, states that Horizon, the NSF Leadership-Class Computing Facility system at UT Austin, uses a shared VAST file system for $HOME and $SCRATCH. These file systems are not Lustre and do not support stripe count or stripe size settings. $WORK (VAST) is unavailable to early users and planned for late 2026. $SCRATCH has no quota, about 400 PB overall capacity, no backup and a 10-day access-time purge. A notice dated 07/24/2026 still limits Horizon to internal users.","sledRelevance":"Direct higher-education research relevance: the largest NSF academic system documents VAST as its primary user file system, giving public universities and state research consortia an operator-written reference for policies, quotas and migration from Lustre. Limited: early-access documentation, not a performance or reliability evaluation.","evidence":"Opened the TACC documentation page; Horizon section shows 'Last update: November 10, 2026' and HTTP Last-Modified 10 Nov 2026 17:37 GMT. File-system table, notices and purge policy read directly. The specific edit made on Nov 10 cannot be identified from the page.","architectureImplications":"Operator design uses one shared VAST namespace for home and scratch over InfiniBand to Grace Blackwell and Vera nodes, without Lustre striping controls; a third $WORK tier on VAST is staged later. Workflows tuned for Lustre stripe settings must be re-validated.","governanceImplications":"Data-retention rules are operator policy, not VAST features: scratch is unbacked with 10-day purge, and deliberately altering access times to evade purge is prohibited. Institutions copying this model need written retention and backup policies.","securityPrivacyImplications":"Not addressed for VAST specifically; the guide covers access via TACC MFA and allocations but says nothing about tenancy isolation or encryption on the VAST file systems.","caveats":"Draft documentation; internal inconsistency between 'up to 200PB' local solid-state storage in the specification table and '~400 PB' scratch capacity in the file-system table; quotas listed as TBD; no measured throughput, uptime or user outcomes; operator rather than independent evaluator.","streamIds":["vast-data"],"roles":{"sales":"Framing: this is an operator reference, not a sales proof point. TACC's public user guide shows a flagship NSF academic system running home and scratch on a shared VAST file system, which helps university research-computing leaders picture an all-flash, non-Lustre user environment. Use it to open conversations about migration effort, purge and backup policy, and staged file-system rollout. Do not cite it as evidence of performance, reliability or cost savings, because the guide is an early-access draft with no measurements and inconsistent capacity figures.","engineering":"Framing: the engineering signal is a non-Lustre user environment at scale. Horizon exposes VAST for $HOME and $SCRATCH over InfiniBand, with no stripe count or stripe size controls, and defers $WORK to late 2026. Engineers advising a campus or state consortium should inventory Lustre-specific tuning in user workflows, test metadata-heavy and small-file patterns, and confirm how quotas and QoS will be enforced. The guide's 200 PB versus about 400 PB discrepancy must be resolved with TACC before any sizing comparison is drawn.","delivery":"Framing: delivery lessons are policy and staging, not product features. TACC opens Horizon to early users with two VAST file systems, a 10-day scratch purge, no scratch backup and a third tier deferred. A university replicating this model needs a user-migration plan, written retention and backup rules, purge communication, and training for researchers leaving Lustre habits behind. Acceptance should be defined against measured user throughput and support tickets during early operations, since the guide records none of these outcomes and remains a draft."},"roleDetails":{"sales":{"customerProblem":"Research universities replacing aging Lustre-based HPC systems need a credible operator example of an all-flash shared file system for home and scratch.","stakeholders":["Research computing director","CIO or CTO","Vice president for research","HPC user advisory committee","Procurement office"],"discoveryQuestions":["Which current workflows depend on Lustre striping or Lustre-specific tools?","What scratch purge and backup policies do researchers accept today?","Is the institution planning a staged rollout with early users before a work or project tier?"],"valueHypothesis":"Interpretation: a single shared VAST namespace may simplify user file-system administration compared with tuned Lustre; this guide documents the design only and does not measure that benefit.","potentialEngagement":"A bounded file-system migration readiness review comparing an institution's Lustre-dependent workflows with the Horizon-style VAST home/scratch model.","unsupportedClaims":["Any performance, throughput or uptime figure for Horizon's VAST storage","Cost or operational savings versus Lustre","Final production capacity, given the 200 PB versus ~400 PB inconsistency"]},"engineering":{"fit":"Fits large shared research-computing environments where home and scratch can move off Lustre and users can live without stripe controls.","architectureIntegration":"Shared VAST file system mounted for $HOME and $SCRATCH over InfiniBand to GPU and CPU node classes, with $WORK added later as a separate VAST tier.","prerequisites":["Inventory of Lustre-specific job tuning (lfs setstripe and similar)","High-bandwidth fabric to compute nodes","Defined quota and purge policy"],"constraints":["No stripe count or stripe size control","$WORK unavailable during early access","Quotas still TBD in the guide","Single U.S. NSF facility; not a state or K-12 deployment"],"security":"Guide documents MFA-based access to TACC systems but not VAST-level tenancy isolation, encryption or audit; these require separate review.","proofOfValue":["Benchmark representative metadata-heavy and small-file jobs without striping","Confirm actual usable scratch capacity against the 200 PB and ~400 PB figures","Test QoS behavior under mixed GPU and CPU job loads"]},"delivery":{"work":"Stage user file systems on VAST (home and scratch first, work later), publish retention and purge rules, and migrate researchers from Lustre habits.","dependencies":["Completed facility power and fabric","Early-user allocation process","Documentation and user-news channel"],"ownership":"University research-computing center operations team, with the vendor for storage support.","skillsAdoption":"Researchers and support staff need training on non-Lustre behavior, purge policy and data-management practices before production.","governanceCheckpoints":["Approve scratch purge and no-backup policy","Approve quotas before general availability","Review readiness before opening the $WORK tier"],"acceptanceCriteria":["Proposed: early-user jobs meet agreed I/O throughput targets on representative workloads","Proposed: documented quotas replace TBD entries before production","Proposed: purge and backup policies published and acknowledged by users"],"risks":["Workflows relying on striping may underperform or fail","Capacity expectations may be wrong if documentation figures stay inconsistent","Data loss from purge if users misunderstand unbacked scratch"]}},"retrievedAt":"2026-11-10T22:09:40-06:00","enrichedAt":"2026-11-10T22:13:45-06:00","enrichmentBasis":"retrieved source","accessibilityWorkforceImplications":"Researchers moving from Frontera or other Lustre systems must relearn file-system practices; the draft guide and a user-news subscription are the documented support channels.","procurementImplications":"Institutions writing HPC storage RFPs can reference a public operator design, but should require measured acceptance tests rather than citing Horizon as proof of performance.","operatingModelImplications":"Operating model separates storage platform (VAST) from operator policy (purge, backup, quotas, staged tiers), with the center owning user-facing policy and documentation.","peerReview":"unknown","sourceVerification":{"openedUrl":"https://docs.tacc.utexas.edu/hpc/horizon/","referenceExcerpt":"Horizon will use a shared VAST file system for the $HOME and $SCRATCH directories.","promptVersion":"sled-research-v4.0","model":null,"basis":"agent-reported inspection"}}}]}