CloudLab

cloudlab.us · Compute, Storage · 3 records · JSON

CloudLab Computefails: identityinstitutionno cardDwrite-gated
CloudLab c240g1 / per-node local blockstore (40 TB) Storagefails: identityinstitutionno cardR
CloudLab persistent remote dataset (10 TB, iSCSI) Storagefails: identityinstitutionno cardR

CloudLab Compute

fails: identityinstitutionno cardDwrite-gated

docs.cloudlab.us/users.html · GPU cloud · compute/cloudlab · JSON · compare

Free tier	CloudLab gives free bare-metal nodes with no allocation accounting at all: "Registering doesn't cost anything, it's simply for accountability. We just ask that if you're going to use CloudLab for anything other than light use, you tell us a bit more about who you are." The Utah cluster alone is 9,336 cores across 1,000-plus nodes, including 28 nodes of 32-core AMD EPYC 7452 with 100Gb Ethernet and 36 nodes of 24-core EPYC 7402P with 25/100Gb links. unit GiB; metered on wall-clock; unmetered-bare-metal
Price	none stated
Per-object cap	not applicable
Retention	not applicable
Egress	not stated
Identity	institution not published
Limits	write-gated Creating a project requires being "faculty, senior research staff, or in some other senior position" and passes a staff review that takes a few days; joining one requires the leader's approval, so an unaffiliated person cannot obtain nodes on their own. source Creating a project requires being "faculty, senior research staff, or in some other senior position" and passes a staff review that takes a few days; joining one requires the leader's approval, so an unaffiliated person cannot obtain nodes on their own.
Binding	write-gated Creating a project requires being "faculty, senior research staff, or in some other senior position" and passes a staff review that takes a few days; joining one requires the leader's approval, so an unaffiliated person cannot obtain nodes on their own.
Interface	unknown inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	The absence of a quota is the unusual part and it is what makes CloudLab usable for a long-running measurement rather than a benchmark: a project can hold nodes for days provided the profile reserves them, which is the shape a network or storage experiment needs and which no credit-based programme provides.
Reliability	no reliability event recorded
Notes	This is the only provider with no allocation counter, so the constraint is purely the project-leader gate plus node availability. The GPU caveat matters for this corpus: the four expandable d750 nodes exist but are withheld from general use, so a record that claimed free GPUs at CloudLab would be wrong. Disk and RAM per node are not on the page read and are left null rather than estimated. The corpus holds no CloudLab record.
Evidence rung	documented
Checked	2026-10-22 (live); stale on 2026-11-21

Evidence (5)

date	source	what it proved
2026-10-22	docs.cloudlab.us/users.html	"Registering doesn't cost anything, it's simply for accountability", quoted
2026-10-22	docs.cloudlab.us/users.html	"We therefore require that project leaders be faculty, senior research staff, or others who are relatively senior positions"; "You should only start a new project if you are a faculty member, senior research staff, or in some other senior position", quoted
2026-10-22	docs.cloudlab.us/users.html	a new project application "will be reviewed by our staff" and "The review process may take a few days", quoted
2026-10-22	docs.cloudlab.us/hardware.html	Utah cluster is 9,336 cores; d6515 is 28 nodes of 32-core AMD 7452 with 100Gb Ethernet; c6525-100g is 36 nodes of 24-core AMD 7402P with 25/100Gb, quoted
2026-10-22	docs.cloudlab.us/hardware.html	the four GPU-capable d750 nodes are "not available for general use. CloudLab staff will grant access to your project on request", quoted

More fields (1)

{
 "wave2": {
  "class": "open-science-grid",
  "free_tier_kind": "unmetered-bare-metal",
  "shape": {
   "vcpu": 32,
   "ram_gib": null,
   "gpu_class": null,
   "gpu_count": null
  }
 }
}

Unknown or withheld values (10)

$.egress.metered egress-unknown
$.egress.terms not-stated
$.identity_cost.detail process
$.interface interface-unknown
$.per_object_cap.notes process
$.per_object_cap.stated not-applicable
$.reliability_note none-observed
$.retention.notes process
$.retention.stated not-applicable
$.tooling not-stated

CloudLab c240g1 / per-node local blockstore (40 TB) Storage

fails: identityinstitutionno cardR

docs.cloudlab.us/advanced-storage.html · research testbed - free bare metal with local HDDs · storage/cloudlab-c240g1-per-node-local-blockstore · JSON · compare

Free tier	c240g1 node type = 12 x 3 TB HDDs (36 TB raw) on 5 nodes at Wisconsin; the storage table prices a 'per-node local blockstore' at 90GB-40TB, so a multi-TB blockstore is requestable by size in the profile and the system picks a node type unit TB; metered on stored-bytes
Price	none stated
Per-object cap	not stated
Retention	Default experiment lifetime is 'a few hours'; extensions need a written justification reviewed by staff, indefinite needs the steering committee
Egress	free - nodes are on 10-25 Gb experiment fabric plus a 1-25 Gb control link to the internet; no byte meter of any kind
Identity	institution institution
Limits	write-gated reviewed by staff, indefinite needs the steering committee source local datasets are ALWAYS ephemeral - node disks are re-imaged after every experiment use, and the manual is explicit that infrastructure is 'not a storage provider' and node disks are 'not configure[d] in any redundant way'. Default experiment lifetime is 'a few hours'; extensions need a written justification reviewed by staff, indefinite needs the steering committee. Usage is metered in node-hours, not bytes. The Emulab parent facility has BYTE-IDENTICAL terms and shares the account: 'Emulab is a public facility, available without charge to most researchers worldwide', but project leaders must still be 'faculty, senior research staff, or others who are relatively senior positions', and an Emulab login is accepted directly as a CloudLab login - so it is a second door onto the same hardware with the same institutional lock.
Binding	none the quota is real and unreachable: identity is the binding constraint, and it is not one of the six limits kinds
Interface	s3, webdav, scp, rsync, ssh ssh (key-only) + root/sudo on real hardware; anything you install - S3, WebDAV, rsync, nginx
Tooling	geni-lib RSpec profiles, portal API token, terraform-provider-cloudlab, ssh/scp
Composes with	run rclone/S3/WebDAV server directly on the node; put Cloudflare in front of it and the fabric becomes the origin. This is the only path in the whole research where the origin has no provider and no byte bill.
Reliability	No redundancy on node disks by design; a single drive failure loses the blockstore, and the whole volume is destroyed at experiment termination.
Notes	c240g1 node type = 12 x 3 TB HDDs (36 TB raw) on 5 nodes at Wisconsin; the storage table prices a 'per-node local blockstore' at 90GB-40TB, so a multi-TB blockstore is requestable by size in the profile and the system picks a node type
Evidence rung	reported One page read moves it to documented.
Checked	2026-10-21 (inherited); stale on 2026-11-04

Evidence (6)

date	source	what it proved
2026-10-21	docs.cloudlab.us/hardware.html	c240g1 = 5 nodes, Haswell 16 core, 128GB RAM, two 480GB SATA SSD + 'Twelve 3 TB HDDs donated by Seagate'; c240g2 = 3 nodes with four 3TB HDDs
2026-10-21	docs.cloudlab.us/advanced-storage.html	storage table: 'Per-node local blockstore \| No [persistent] \| 90GB to 40TB \| 100-1000MB/sec \| When up to 40TB is needed; specified by size in profile, system picks the node type'
2026-10-21	docs.cloudlab.us/basic-concepts.html	'All experiments have an expiration time. By default, the expiration time is short (a few hours)' and longer extensions 'require the intervention of CloudLab staff'
2026-10-21	www.cloudlab.us/aup.php	Use for research and educational purposes is allowed'; litmus test is intent to publish in scholarly venues
2026-10-21	docs.emulab.net/	the parent facility shares the account: 'Emulab is a public facility, available without charge to most researchers worldwide'
2026-10-21	docs.emulab.net/users.html	'We therefore require that project leaders be faculty, senior research staff, or others who are relatively senior positions'

More fields (1)

{
 "dead_end": {
  "failure_class": "quota",
  "why": "c240g1 node type = 12 x 3 TB HDDs (36 TB raw) on 5 nodes at Wisconsin; the storage table prices a 'per-node local blockstore' at 90GB-40TB, so a multi-TB blockstore is requestable by size in the profile and the system picks a node type",
  "retry_when": null,
  "first_recorded": "2026-10-21"
 }
}

Unknown or withheld values (5)

$.extras.dead_end.retry_when process
$.free_tier.notes process
$.per_object_cap.notes not-extracted
$.per_object_cap.stated not-extracted
$.retention.notes not-extracted

CloudLab persistent remote dataset (10 TB, iSCSI) Storage

fails: identityinstitutionno cardR

docs.cloudlab.us/advanced-storage.html · research testbed - persistent block storage · storage/cloudlab-persistent-remote-dataset · JSON · compare

Free tier	up to 10 TB per persistent remote dataset; pool is 500 TB at Utah, 160 TB at Wisconsin, 40 TB at Clemson. Anything over 1 TB 'likely to require administrative approval'; long-term datasets are 'subject to per-project quotas for total size and might require interaction with CloudLab staff' unit TB; metered on unstated
Price	none stated
Per-object cap	10 TB hard ceiling per (10000000000000 bytes)
Retention	not stated
Egress	free; mounted as /dev/sdb via iSCSI on the experiment network, 50-200 MB/s
Identity	institution institution
Limits	write-gated approval' source per-object-cap 10 TB hard ceiling per source 10 TB hard ceiling per dataset, so 20 TiB needs two datasets and two projects' worth of quota. >1 TB needs a human to approve. Simultaneous read-write sharing across nodes 'will almost certainly wind up with a corrupted dataset'. No off-site backup; RAID/ZFS on the storage server only.
Binding	none the quota is real and unreachable: identity is the binding constraint, and it is not one of the six limits kinds
Interface	unknown block device -> any filesystem -> any server you run on the node
Tooling	portal 'Create Dataset' UI, RemoteBlockstore in RSpec, rwclone/readonly mapping for multi-node
Composes with	persistent dataset + one node running S3/WebDAV + Cloudflare in front = 10 TB of genuinely durable-by-dint-of-not-being-wiped storage at $0
Reliability	Storage servers have RAID/ZFS but 'most clusters do not have off-site backup of user data'; a project expiry deletes the allocation.
Notes	up to 10 TB per persistent remote dataset; pool is 500 TB at Utah, 160 TB at Wisconsin, 40 TB at Clemson. Anything over 1 TB 'likely to require administrative approval'; long-term datasets are 'subject to per-project quotas for total size and might require interaction with CloudLab staff'
Evidence rung	reported One page read moves it to documented.
Checked	2026-10-21 (inherited); stale on 2026-11-04
Parsed numbers	`$.limits[1].value` limit:per-object-cap 10.0 TB

Evidence (2)

date	source	what it proved
2026-10-21	docs.cloudlab.us/advanced-storage.html	'Persistent remote datasets ... Capacity: up to 10TB'; 'sizes in excess of 1TB are likely to require administrative approval'; 'long-term datasets ... subject to per-project quotas for total size and might require interaction with CloudLab staff'
2026-10-21	docs.cloudlab.us/hardware.html	'The Utah Cloudlab cluster includes a storage server for remote datasets. The server currently has 500TB available for allocation'; Wisconsin 160TB; Clemson 40TB

More fields (1)

{
 "dead_end": {
  "failure_class": "write-gated",
  "why": "up to 10 TB per persistent remote dataset; pool is 500 TB at Utah, 160 TB at Wisconsin, 40 TB at Clemson. Anything over 1 TB 'likely to require administrative approval'; long-term datasets are 'subject to per-project quotas for total size and might require interaction with CloudLab staff'",
  "retry_when": null,
  "first_recorded": "2026-10-21"
 }
}

Unknown or withheld values (6)

$.extras.dead_end.retry_when process
$.free_tier.notes process
$.interface interface-unknown
$.per_object_cap.notes not-extracted
$.retention.notes not-extracted
$.retention.stated not-extracted