Frank Würthwein - NRP and the Path forward

NRP & the Path forward
Frank Würthwein
Director, San Diego Supercomputer Center
February 9th 2023

NATIONAL RESEARCH
3
https://mindsweneed.org
Democratize Access

NATIONAL RESEARCH
NATIONAL RESEARCH
Long Term Vision
• Create an Open National Cyberinfrastructure that
allows the federation of CI at all ~4,000 accredited,
degree granting higher education institutions, non-
profit research institutions, and national laboratories.
§ Open Science
§ Open Data
§ Open Source
§ Open Infrastructure
4
Openness for an Open Society
Open Compute
Open Storage & CDN
Open devices/instruments/IoT, …?

NATIONAL RESEARCH
NATIONAL RESEARCH
Community vs Funded Projects
5
Community with
Shared Vision
Lot’s of funded projects that
contribute to this shared
vision in different ways.
We want you to …
… grow NRP.
… build on NRP.
NRP is “owned” and “built” by the community for the community

How is NRP different from OSG?
We’ve been doing federated cyberinfrastructure since 2005.
Why is NRP even needed given that OSG exists?

NATIONAL RESEARCH
HTCondor/OSG
Cyberinfrastructure Stack
7
Hardware
IPMI, Firmware, BIOS
Kubernetes
Admiralty
SLURM
NRP operates at all layers of the stack, from IPMI up
• IPMI reduces TCO and lower threshold to entry
• Kubernetes allows service deployments
• Also the natural layer for application container deployment
• Admiralty allows K8S federation with folks who want control
• Including cloud integration to access TPUs & other cloud only architectures
• HTCondor allows NRP to show up as a “site” in OSG
The layer you integrate at depends on
• Control you want
• Effort you can afford

Complementarity in Implementation of
“Bring Your Own Resource” model
OSG/PATh focused on campus cluster integration.
NRP focused on individual node integration instead of clusters.

NATIONAL RESEARCH
HTCondor/OSG
9
Hardware
Kubernetes
Admiralty
SLURM
• IPMI reduces TCO and lowers threshold to entry
• HTCondor allows NRP to show up as site in OSG
• Under-resourced institutions
• Network providers and their POPs
• CS & ECE faculty specialized on:
• AI/ML => gaming GPUs
• systems R&D
All of these find it difficult to
justify staff to support all layers

NATIONAL RESEARCH
Hardware on NRP is Global
10
NRP integrates hardware in USA, EU, and Asia

NATIONAL RESEARCH
11
Hardware
Kubernetes
• IPMI reduces TCO and lowers threshold to entry
• HTCondor allows NRP to show up as site in OSG
• Open Science Data Federation
• Origins & Caches in US, EU, Asia
• Protein Data Bank
• (Future) Replicas in EU & Asia
NRP is unique in its support of
global service deployments
OSDF ops by PATh PDB ops by PDB ???

Supporting Nautilus
for the next decade
Nautilus = K8S infrastructure of PRP for the last 5+ years
Nautilus = K8S of NRP for the next 10 years

NATIONAL RESEARCH
NATIONAL RESEARCH
The NSF Cat-II Program
• NSF supports via the Cat-II program novel systems ideas.
- 3 year “testbed” phase
§ The PI owns the resource, and has (some) freedom regarding who uses it.
Ÿ No requirements for making it available via any specific allocation mechanism.
§ It is expected that not all features work on day 1.
Ÿ 3 years of experimentation & development of features
- 2 year “allocation” phase
§ The resource is made available via an NSF supported allocation mechanism.
- The solicitation mentions the possibility of an additional 5 year
renewal without re-competition if system is successful.
• We decided that this is an ideal program to try and secure
NRP core operations funding for the next 10 years
- And thus provide the stability necessary for growth of NRP.
13

NATIONAL RESEARCH
NATIONAL RESEARCH
Cat-II: Prototype National
Research Platform (PNRP)
14
Funded as NSF 2112167
80GB A100
A10
64
288
5 year project with $5M hardware & $6.45M people
Supports Nautilus, and thus the core NRP infrastructure
Promises to build on “PRP” functionality, and go beyond
NSF Acceptance Review of System scheduled for March 8 & 9th 2023
PI = Wuerthwein; Co-PIs: DeFanti, Rosing, Tatineni, Weitzel

NATIONAL RESEARCH
NATIONAL RESEARCH
Innovations
• I1: Innovative network fabric that allows “rack” of hardware to
behave like a single “node” connected via PCIe.
• I2: Innovative application libraries to expose FPGAs hardware
to science apps at language constructs scientists understand
(C, C++ rather than firmware)
• I3: A “Bring Your Own Resource” model that allows campuses
nationwide to join their resources to the system.
• I4: Innovative scheduling to support urgent computing, including
interactive via Jupyter.
• I5: Innovative Data Infrastructure, including national scale
Content Delivery System like YouTube for science.
15
I3 & I4 & I5 turn ”PRP” into “NRP” and sustains it into the future.
I1 & I2 are totally new.

NATIONAL RESEARCH
NATIONAL RESEARCH
Data Infrastructure Model of NRP
• Support regional Ceph storage systems across the USA.
- Campuses can join individual storage hosts to the Ceph system in their region.
- All regional storage systems are Origins in OSG Data Federation (OSDF)
- Deploy replication system such that researchers can decide what part of their
namespace should be in which regional storage.
• Deploy caches in Internet2 backbone such that no campus nationwide is
more than 500 miles from a cache.
16
NRP data infrastructure model combines best of PRP & OSG
From PRP we take the regional Ceph storage concept
From OSG/PATh we take the data origin & caching concepts
And then we add as a totally new feature:
User controlled replication of partial namespaces across regions.
(We will develop this during 3 year “testbed” phase)
Want Others to build higher level data services on top

NATIONAL RESEARCH
NATIONAL RESEARCH
Matrix of Science x Innovations
17
NRP. The networks of researchers we create will use the NRP systems innovations to achieve broader im-
pact, with the help from our collaborators, users, staff, Co-PIs, and PI.
Table 3.1 Representative Science and Engineering Use Cases
Application domain
Lead researcher &
Institution
Science Driver
Themes NRP Innovations
LIGO
Peter Couvares, LIGO Lab;
Erik Katsavounidis, MIT
BGS, UC, AI I2, I3, I4, I5
IceCube Benedikt Riedel, UW Madison BGS, UC, AI I3, I4
Astronomy (DKIST &
Sky Surveys)
Curt Dodds, U. Hawai’i BGS, AI I3, I5,
Campus Scale Instru-
ment Facilities
Mark Ellisman, NCMIR; Sa-
mara Reck-Peterson, Nicon
Imaging Center; Johannes
Schoeneberg, Adaptive Op-
tics Lightsheet Microscopy;
Kristen Jepsen, Institute for
Genomic Medicine; Tami
Brown-Brandl, Precision Ani-
mal Management
SD, UC, H I1, I2, I3, I4, I5
Molecular Dynamics
Rommie Amaro, UCSD; An-
dreas Goetz, SDSC; Jona-
than Allen, LLNL
MD, AI, H I1, I2, I3
Human microbiome Rob Knight, UCSD G, AI, H I1, I2, I3
Genomics & Bioinfor-
matics
Alex Feltus, Clemson G, AI, H I3, I4, I5
Fluid Dynamics Rose Yu, UCSD AI I1, I2, I3
Experimental Particle
Physics, IAIFI Phil Harris, MIT AI, BGS, SD I1, I2
Computer Vision Nuno Vasconcelos, UCSD AI, CV I3
Computer Graphics Robert Twomey, UNL CV, AI I3
Programmable Storage Carlos Malzahn, UCSC SD I1, I2, I5
AI systems software
stack for FPGAs Hadi Esmaeilzadeh , UCSD SD I1, I2
WildFire Analysis &
Prediction
Ilkay Altintas, UCSD UC, AI, CV I3, I4
Key: The NRP innovations column lists those innovations among I1 through I5 listed in Section 2.1 that a
given science driver most benefits from.
NSF MREFCs
Incl. 4 campus scale
instrument facilities
Incl. a very diverse
set of sciences and
engineering
Lot’s of AI …
but so much more …

NATIONAL RESEARCH
NATIONAL RESEARCH
FKW’s Wishlist for the Future
• Growth of NRP infrastructure
- 1,000++ GPUs end of 2022
- 50 PB storage end of 2024
- Growth in diversity of community
§ # and types of campuses and their researchers
• Introduce new capabilities to NRP
- Machine learning at 100TB scale
- Support Domain Specific Architecture R&D on NRP
- Expand NRP into Wireless, Edge, IoT
- Towards “FAIR” on OSDF
• New Directions initiated by the Community
18

NATIONAL RESEARCH
NATIONAL RESEARCH
“Domain Specific Architectures”
• I1: Innovative network fabric allowing “composable hardware”.
• I2: Innovative application libraries allowing “domain optimized
architectures” on FPGAs
19
“end of Moore’s law” motivates new architectures
Mark Papermaster, CTO of AMD
PRISM, a Jump 2.0 project
funded by SRC
is early user of FPGAs@PNRP
John Shalf (2020)
https://doi.org/10.1098/rsta.2019.0061
PI, Tajana Rosing

NATIONAL RESEARCH
NATIONAL RESEARCH
New Data Origins
• The NSF CC* 2022 program awarded 9
campuses with $500k storage awards each.
- We guess this pays for 5PB of storage each.
• Some of these campuses may decide to
integrate their CC* storage into the OSDF.
• Some of these campuses have storage from
other projects that they may integrate with the
OSDF in addition.
• NSF 23-523 includes $500k storage
solicitation again, Spring & Fall 2023.
20

NATIONAL RESEARCH
NATIONAL RESEARCH
Summary & Conclusions
• PRP ended, and was replaced by NRP
- Significant new capabilities via Cat-II system “PNRP”
§ PNRP provides ops effort for Nautilus for the future
- # of GPUs available double in 2022.
§ new GPUs (A10, 3080, 3090, A100) much more powerful than older GPUs
- # of FPGAs increase from a few to a few dozen in 2022.
- # of caches grow by 50% in 22/23
=> more consistent coverage across USA
- Data volume served expected to grow substantially in 23/24/25.
§ How much? As yet too hard to predict.
• Hoping to recruit new partners to build FAIR capabilities on
top of OSDF within the next 5 years.
• Hoping to expand NRP into sensor networks using 5G &
6G in the next 10 years.
21

NATIONAL RESEARCH
NATIONAL RESEARCH
Acknowledgements
• This work was partially supported by the NSF
grants OAC-1541349, OAC-1826967, OAC-
2030508, OAC-1841530, OAC-2005369,
OAC-21121167, CISE-1713149, CISE-
2100237, CISE-2120019, OAC-2112167
22

Frank Würthwein - NRP and the Path forward

More Related Content

Similar to Frank Würthwein - NRP and the Path forward

Similar to Frank Würthwein - NRP and the Path forward (20)

More from Larry Smarr

More from Larry Smarr (20)

Recently uploaded

Recently uploaded (20)

Frank Würthwein - NRP and the Path forward