Low-barrier access to cloud computing on ResearchCloud

a case study

Jelle Treep & Dawa Ometto

Introduction

Research Data Management Services @ Utrecht University:

Research IT & Engineering
Yoda
Applications (GitHub)
→ HPC support & access

Introduction

Large increase in demand for HPC.
No longer just for clusters, but also for Virtual Research Environments (VRE)!
At UU, we meet this demand using SURF services:
- Snellius
- ResearchCloud (VRE)
- RCCS contracts

Research Capacity Computing Service

RCCS gives institutes direct access to SURFs compute facilities:

Lowering barriers
Snellius & SURF Research Cloud
Expertise
pay-per-use model

Research Capacity Computing Service

Costs:

1.000 RCCS-credits - 33,14 euro
~0,03 euro per CPU hour on Research Cloud
~0,025 euro per CPU hour on Snellius
Access to commercial cloud (varying cost)

How do we organize access?

Intake with engineer (30 minutes)
Starting budget of 10.000 credits
→ Onboarding time ~2 days

Support

Support provided by Research Engineers:

Getting started workshop (SRC)
Catalog development (SRC)
Basic user support (Snellius and SRC)
Weekly walk-in hours

History

RCCS @ UU runs for ~8 years (Kees van Eijden)
From ~10 users to >100 users
- 1 every 2 months to 5 per week (peak)

RCCS reports

Research Cloud projects

RCCS reports

Faculty	Projects
Geosciences	16
Science	28
Humanities	20
Social Sciences	17
Law and Economics	13
Veterinary Medicine	13

CPU vs. GPU on Research Cloud

RCCS reports

Source: NWO

What do we learn from this?

Large and growing demand for flexible HPC solutions
- Small-medium (VRE) as well as large (Snellius) scale.
- Across all faculties.
RCCS starting budgets help researchers to conduct exploratory studies, often leading to larger projects.
- 15 million credits granted after starting budget! (in 2023)

What do we learn from this?

Success factors at UU:

Close cooperation with SURF
Low-barrier onboarding & support
Engineering support
- Reliable and reusable VRE environments
- Hands-on customization
Good relations with IT groups at faculties

What do we learn from this?

Most importantly:

Support team of engineers, developers, admins
- since end of 2023: 1 dedicated engineer

Future challenges

Challenges when scaling up further:

Ensuring enough user support and intake capacity, together with faculties
Meeting needs for custom engineering
Billing recurring large users internally

Links