Bring compute to the data
Analyze large datasets where they live, without downloading them
NSF Awards: 1925001 & 2505560
A CloudBank workshop for Rice University research IT
Jose Cols, Rob Fatland, Anthony Arendt
UW eScience Institute · CloudBank
February 2027
Part 1
What the commercial cloud adds to research and teaching
Alongside your HPC cluster, the cloud lets researchers:
Analyze large datasets where they live, without downloading them
Rent GPUs or large-memory machines, and pay only while they run
Many simulations, parameter sweeps or fine‑tuning runs at once
Call hosted models through an API, or serve your own on GPUs
Collaborators and students work in the same setup
Web apps, APIs and live model endpoints that others can reach
The friction
I couldn't get a cloud account on my grant without a university credit card.
I just don't have enough time to learn all this. Can you give me a jump start?
Setting up my first experiment in the cloud took forever, and it was different for every platform I tried.
I got a $12K bill last month. It's opaque, and I don't have time to wade through it.
Paraphrased from conversations with researchers
with an ACCESS or NAIRR Pilot award
Commercial clouds
where your work runs
Run by UC San Diego (SDSC), UC Berkeley and UW eScience · funded by NSF since 2019
Sources: SDSC news, April 2025 · NSF news · ACCESS: CloudBank Research
NSF pays through your ACCESS or NAIRR Pilot award; no purchase order
One account across all the clouds CloudBank offers
Alerts as you spend, and tips on idle resources
Also pays for marketplace software and AI model access
Your cloud accounts are created after onboarding
Your ACCESS account signs you in to each cloud
Add or remove people in one place
Add another cloud to your project later
| For | Research projects of any size | Courses taught with Jupyter notebooks |
|---|---|---|
| You get | Your own accounts on AWS, Google Cloud, Azure and IBM Cloud | A ready-made JupyterHub with grading tools; students just log in from a browser |
| To start | Apply for an ACCESS or NAIRR Pilot allocation | Apply for an Explore ACCESS allocation; CloudBank sets up your hub |
Sources: ACCESS: CloudBank Research · ACCESS: CloudBank Classroom
Part 2
The conversation we have with every first-time cloud user
Is the cloud the right tool?
Nocampus HPC, or ACCESS and NAIRR on-premises resources
YesIs CloudBank the right route?
Noa departmental or institutional cloud account
YesAllocation
ACCESS (Explore, Discover, Accelerate) or the NAIRR Pilot
Onboard
CloudBank opens accounts on AWS, Google Cloud, Azure or IBM Cloud
Work
build, run and watch spending
Exit
archive, publish and close accounts
Use on-premises ACCESS or NAIRR resources instead
Sources: ACCESS allocations policy · CloudBank recommended uses
| LLM task | Cluster | Cloud | Watch out for |
|---|---|---|---|
| Call a hosted model | Not needed | API, billed per use | Data leaves campus |
| Run an open model | Good for batch jobs | Good for always-on services | GPU queues and quotas |
| Fine-tune or train | Best for large runs | Good for small fine-tunes | GPU hours add up fast |
| Tier | Range | Average | Typical use |
|---|---|---|---|
| Explore | up to $10,000 | $5,000 | Grad students, first-time users; 1-page proposal |
| Discover | $10,001–$37,500 | $18,750 | Multi-user app or scaling up; 3-page proposal |
| Accelerate | $37,501–$75,000 | $56,250 | Established gateway, many users, multiple years |
A second way to a CloudBank account, for AI research
Teaching? Request a NAIRR Classroom project for a CloudBank JupyterHub.
Sources: NAIRR Pilot resource requests · CloudBank: NAIRR Pilot Research · NAIRR: CloudBank onboarding · CloudBank: NAIRR Pilot Classroom
Use the cloud cost calculators to estimate expenses.
Cloud providerCreate an ACCESS account and request credits.
ACCESSTurn ACCESS credits into CloudBank dollars.
ACCESSFollow the instructions in your onboarding email.
CloudBankTags show where each step happens. Cloud providers: AWS, Google, Azure, IBM.
Awards end on a date; nothing shuts itself down when they do.
Part 3
A live walkthrough: account, use case and AI-assisted setup
Treat its answer as a draft: run it, read the errors, ask again.
I have an AWS account through CloudBank.
Give me AWS CLI steps to run a Python API
on a small spot instance in us-east-1.
Tag everything project=rice-demo and stop
the instance automatically after 4 hours.
Assistants often skip tags and auto-stop unless you ask for them.
Activity
6 min
Deliverable
A working endpoint, then a stopped one
Part 4
Design around storage, compute and network, then choose the abstraction
Answer these before choosing a service; the cost follows from the answers.
vLLM on a GPU node
A managed model endpoint
A chat API
Same task each time: more control and more responsibility at every step.
from openai import OpenAI
client = OpenAI(base_url=URL, api_key=KEY)
reply = client.chat.completions.create(
model=MODEL,
messages=[{"role": "user", "content": q}])
Point URL at a hosted API or at your own vLLM server; the rest stays.
Source: vLLM documentation, docs.vllm.ai
resources:
accelerators: A10G:1
ports: 8000
setup: pip install vllm
run: vllm serve $MODEL --port 8000
sky launch finds an A10G, installs vLLM and serves the model on port 8000.
Source: SkyPilot documentation, docs.skypilot.co
Go deeper with the take-home tutorials and the Cloud Clinic recordings.
CloudBank credits let you build on a real account, not a sandbox.
Activity
2 weeks
Deliverable
A one-page note for your researchers
Part 5
And how research IT can head them off
| Pitfall | What it looks like | What research IT can do |
|---|---|---|
| Credits never used | Award granted, nothing launched | Book an onboarding call in week one |
| Learning curve underestimated | Months before a first result | Offer a starter template and office hours |
| Zombie instances | Bills grow after the work ends | Require tags, alerts and autostop |
| Leaked access keys | Unknown instances appear overnight | Use SSO and short-lived credentials |
| Wrong service | A big VM doing a managed service's job | Review the design before launch |
| Wrong storage | Hot storage for archives; heavy egress | Set lifecycle rules; compute in-region |
What common resources cost, to the nearest round number
Left on for a month, a $1/hour machine costs about $730. Turn it off when you are done.
Prices: AWS on-demand, US East (N. Virginia), October 2026. Amazon S3 pricing · Amazon EC2 on-demand pricing
Activity
3 min
Deliverable
One guardrail per scenario, shared aloud
Part 6
From single projects to a marketplace of research resources
Awards from October 2025 to October 2026
| Tier | Projects | Average |
|---|---|---|
| Explore | 234 | $4,423 |
| Discover | 56 | $19,689 |
| Accelerate | 2 | $52,094 |
plus 40 more fields
Source: CloudBank allocation records, October 2025 to October 2026
Open science is the exit plan that keeps paying off.
Activity
3 min
Deliverable
A shortlist of Rice resources worth sharing
Learning & support
Cloud 101 basics, plus a new Cloud Clinic deep dive about every month
cloudbank.org/video‑tutorialsCompare compute, storage and data services across the clouds
cloudbank.org/cloudbank‑catalogMonthly on Zoom, 10–11 am Pacific. Bring a draft proposal or a billing question.
Write to us help@cloudbank.org
Help us improve this workshop — your response takes about 2 minutes.
Apply for credits and try one example at each level.