1. Global Network Advancement Group
Next Generation Network-Integrated Systems
4th National Research Platform Workshop:
NRP, GRP, GNA-G Partnership
Harvey Newman, Caltech, GNA DIS WG Chair
February 9, 2023

2. Higgs Boson
Couples to Mass
Ten Year Anniversary of the Higgs Boson Discovery: July 4 2022
Search for Higgs
Boson Pairs
A portrait of the Higgs boson by the CMS experiment ten years after the discovery | Nature

3. 3
10th Anniversary of the Higgs Boson Discovery
and Beyond; 75 Years of Exploration !
Advanced Networks Were Essential
to Higgs Discovery and Every Ph.D
Thesis; They will be Essential
to All Future Discoveries
•NOTE: ~95% of Data Still to be Taken
•Greater Intensity: Upgraded
detectors for more complex events
•To 5X Data Taking Rate in 2029-40
Englert
2013 Nobel Prize
Higgs
48 Year Search; 75 Year Exploration
Theory (1964): 1950s – 1970s;
LHC + Experiments Concept: 1984
Construction: 2001; Operation: 2009
Run1: Higgs Boson Discovery 2012
Run2 and Going Forward:
Precision Measurements and BSM
Exploration: 2013 - 2042


4. Global Data Flow: A Worldwide System Conceived
by Caltech and UCSD. First Tier2 in 1999-2001
Tier 1
Tier2 Center
Online System
CERN Center
100 PBs of Disk;
Tape Robot
Chicago
Institute
Institute
Institute
Institute
Workstations
~1000
MBytes/sec
10 to 100 Gbps
Physics data
cache
~PByte/sec
N X100 Gbps
Tier2 Center
Tier2 Center
Tier2 Center
10 to 100 Gbps
Tier 0
Tier 3
Tier 4
Tier 2
Experiment
London Paris Taipei
CACR
13 Tier1, 170 Tier2
+ 300 Tier3 Centers:
1k to N X 10k Cores
 Tier0: Real Time Data Processing
Tier1 and Tier2: Data and Simulation
Production Processing
Tier2 and Tier3: Data Analysis
Increased Use as a Cloud Resource (Any Job Anywhere)
Increased “Elastic” Use of Additional HPC and Cloud Resources
A Global Dynamic System: Fertile Ground for Control with ML
Bologna

5. Core of LHC Networking LHCOPN,
LHCONE, GEANT, ESnet, Internet2, CENIC…
+ NRENs in Europe, Asia, Latin America, Au/NZ; US State Networks
LHCOPN: Simple & Reliable
Tier0+1 Ops GEANT
Internet2
ESnet (with EEX) CENIC and NRP
LHCONE VRF: 170 Tier2s ++

6. + NRENs in Europe, Asia, Latin America, Au/NZ; US State Networks
LHCOPN: Simple & Reliable
Tier0+1 Ops GEANT
Internet2
ESnet (with EEX) CENIC and NRP
LHCONE VRF: 170 Tier2s
CENIC and Pacific Wave: A “Regional” Network for California
with Global Reach and Vision East and West

7. WLCG Transfers Dashboard: 3/2017 to 3/2022
HL LHC “10%” Data Challenge in October 2021 exceeded 100 Gbytes/sec for >1 Day
The “30%” Challenge Scheduled for 2024 will be a real challenge
General WLCG Traffic is again at or above the pre-pandemic level
https://monit-grafana.cern.ch/d/AfdonIvGk/wlcg-transfers?orgId=20

8. Towards a Computing Model for the HL LHC Era
Challenges: Capacity in the Core and at the Edges
 Programs such as the LHC have experienced rapid exponential traffic growth,
at the level of 40-60% per year
 At the January 2020 LHCONE/LHCOPN meeting at CERN, CMS and ATLAS
expressed the need for Terabit/sec links on major routes,
by the start of the HL-LHC in ~2029
 This is projected to outstrip the affordable capacity
 The needs are further specified in “blueprint” Requirements documents
by US CMS and US ATLAS, submitted to the ESnet Requirements Review in
August 2020, and captured in a comprehensive DOE Requirements Review
report for HEP: https://escholarship.org/uc/item/78j3c9v4
 Three areas of particular capacity-concern by 2028-9 were identified:
(1) Exceeding the capacity across oceans, notably the Atlantic,
served by the Advanced North Atlantic (ANA) network consortium
(2) Tier2 centers at universities requiring 100G 24 X 7 X 365 average
throughput with sustained 400G bursts (a petabyte in a shift), and
(3) Terabit/sec links to labs and HPC centers (and edge systems)
to support multi-petabyte transactions in hours rather than days 8

9. Mission: To Support global research and education using the
technology, infrastructures and investments of its participants
The GNA-G exists to bring together researchers, National Research and Education
Networks (NRENs), Global eXchange Point (GXP) operators, regionals and other R&E
providers, in developing a common global infrastructure to support the needs.
https://www.gna-g.net/

10. GREN: Collaboration on the intercontinental transmission layer
Telemetry
Data Intensive
Science
AutoGOLE/SENSE
GREN Mapping
GXP Architectures & Services
Connecting Offshore
Students
Research & Development Operational
Link consortia
APR ANA …
AER AmLight
GNA
Architecture
v2.0
GNA-G Leadership Team
Routing Anomalies Network Automation
GNA-G Executive Liaison
GNA-G participant CEOs etc
Global NREN CEO Forum
Securing the GREN
Strategic Security Focus Group
GNA-G overview
and Vision

11. The GNA-G Data Intensive Sciences WG
 Principal aims of the GNA-G DIS WG:
(1) To meet the needs and address the challenges
faced by major data intensive science programs
 In a manner consistent and compatible with support for the needs of
individuals and smaller groups in the at large A&R communities
(2) To provide a forum for discussion, a framework and shared tools for short
and longer term developments meeting the program and group needs
 To develop a persistent global testbed as a platform, to foster
ongoing developments among the science and network communities
 While sharing and advancing the (new) concepts, tools & systems needed
 Members of the WG partner in joint deployments and/or developments of
generally useful tools and systems that help operate and manage R&E
networks with limited resources across national and regional boundaries
 A special focus of the group is to address the growing demand for
 Network-integrated workflows
 Comprehensive cross-institution data management
 Automation, and
 Federated infrastructures encompassing networking, compute, and storage
 Working Closely with the AutoGOLE/SENSE WG, NRP and GRP
1
1
Charter: https://www.dropbox.com/s/4my5mjl8xd8a3y9/GNA-G_DataIntensiveSciencesWGCharter.docx?dl=0

12. The GNA-G Data Intensive Sciences WG
 Mission: Meet the challenges of globally distributed data and computation
faced by the major science programs
 Coordinate provisioning the feasible capacity across a global footprint,
and enable best use of the infrastructure:
 While meeting the needs of the participating groups, large and small
 In a manner Compatible and Consistent with use by the at-large A&R communities
 Members:
Alberto Santoro, Azher Mughal, Bijan Jabbari, Brian Yang, Buseung Cho, Caio Costa, Carlos Antonio
Ruggiero, Carlyn Ann-Lee, Chin Guok, Ciprian Popoviciu, Dale Carder, David Lange, David Wilde, Dima
Mishin, Edoardo Martelli, Eduardo Revoredo, Eli Dart, Eoin Kenney, Frank Wuerthwein, Frederic Loui,
Harvey Newman, Heidi Morgan, Iara Machado, Inder Monga, Jeferson Souza, Jensen Zhang, Jeonghoon
Moon, Jeronimo Bezerra, Jerry Sobieski, Joao Eduardo Ferreira, Joe Mambretti, John Graham, John Hess,
John Macauley, Julio Ibarra, Justas Balcas, Kai Gao, Karl Newell, Kaushik De, Kevin Sale, Lars Fischer,
Liang Zhang, Mahdi Solemani, Maria Del Carmen Misa Moreira, Marcos Schwarz, Mariam Kiran, Matt
Zekauskas, Michael Stanton, Mike Hildreth, Mike Simpson, Ney Lemke, Phil Demar, Raimondas Sirvinskas,
Richard Hughes-Jones, Rogerio Iope, Sergio Novaes, Shawn McKee, Siju Mammen, Susanne Naegele-
Jackson, Tom de Fanti, Tom Hutton, Tom Lehman, William Johnston, Xi Yang, Y. Richard Yang
 Participating Organizations/Projects:
 ESnet, Nordunet, SURFnet, AARNet, AmLight, KISTI, SANReN, GEANT, RNP, CERN, Internet2,
CENIC/Pacific Wave, StarLight, NetherLight, Southern Light, Pacific Research Platform,
FABRIC, RENATER, ATLAS, CMS, VRO, SKAO, OSG, Caltech, UCSD, Yale, FIU, UERJ,
GridUNESP, Fermilab, Michigan, UT Arlington, George Mason, East Carolina, KAUST
 Working Closely with the AutoGOLE/SENSE WG
 Meets Weekly or Bi-weekly; All are welcome to join.
12
Charter: https://www.dropbox.com/s/4my5mjl8xd8a3y9/GNA-G_DataIntensiveSciencesWGCharter.docx?dl=0

13. AutoGOLE / SENSE Working Group
 Worldwide collaboration of open exchange points and R&E networks
interconnected to deliver network services end-to-end in a fully
automated way. NSI for network connections, SENSE for integration
of End Systems and Domain Science Workflow facing APIs.
 Key Objective:
 The AutoGOLE Infrastructure should be persistent and reliable,
to allow most of the time to be spent on experiments and research.
 Key Work areas:
 Control Plane Monitoring:
Prometheus based, deployment now underway
 Data Plane Verification and Troubleshooting Service:
Study and design group formed
 AutoGOLE related software: Ongoing enhancements to facilitate
deployment and maintenance (Kubernetes, Docker based systems)
 Experiment, Research, Multiple Activity, Use Case support:
Including NOTED, Gradient Graph, P4 Topologies, Named Data
Networking (NDN), Data Transfer Systems integration and testing.
 WG information
https://www.gna-g.net/join-working-group/autogole-sense

14. 14
Towards a Next Generation
Network-Integrated System
For Data Intensive Sciences
The Global Network
Advancement Group
The AutoGOLE/SENSE and
Data Intensive Sciences WGs
Partnering with GRP and NRP

15. GNA-G AutoGOLE/SENSE WG Global Persistent Testbed
Software Driven Network OSes; Programmability at the Edges and in the Core

16. P4 + SONIC Programmable Global PersistentTestbed
19 Active GNA-G/RARE P4, SONIC Testbed Sites
• Caltech, Pasadena-US: 4 FreeRtr/P4 + SONIC
• CERN, Geneva-CH: FreeRtr/P4
• FIU, Miami-US: FreeRtr/P4
• GEANT, Amsterdam-NL: FreeRtr/P4
• GEANT, Budapest-HU: FreeRtr/P4
• GEANT, Frankfurt-DE: FreeRtr/P4
• GEANT, Paris-FR: FreeRtr/DPDK
• GEANT, Poznan-PL: FreeRtr/P4
• GEANT, Prague-CZ: FreeRtr/DPDK
• RENATER, Paris-FR: FreeRtr/P4
• SouthernLight (FIU/RedClara/Rednesp/RNP),
São Paulo-BR: FreeRtr/P4
• StarLight, Chicago-US: FreeRtr/P4
• SWITCH, Geneva-CH: FreeRtr/P4
• TCD, Dublin-IE: FreeRtr/P4
• Tennessee Tech: FreeRtr/P4
+ 9 Expected sites (by SC22):
• HEAnet, Dublin-IE: FreeRtr/P4
• JISC, London-UK: FreeRtr/P4
• KAUST, Saudi Arabia: FreeRtr/DPDK
• KISTI, South Korea: SONiC/P4
• RNP, Rio de Janeiro-BR: FreeRtr/P4,
SONiC/P4
• SC22 Caltech Booth, Dallas-US:
FreeRtr/P4
• UCSD, San Diego-US: SONiC/P4
• UFES, Vitória-BR: 2x FreeRtr/P4
• UMd, College Park, Maryland-US:
FreeRtr/P4

17. GEANT (Now Global) P4 Lab
and Dataplane of P4 Programmable Switches
A new worldwide platform for
 New agile and flexible feature
development
 New use case development
corresponding to research
programs’ requirements
 Multiple research network
overlay slices
A global playing field for
 Next generation network
monitoring tools and systems
 Development of fully automated
network deployment
 New network operations paradigm
development
VISION: Federate and integrate multiple testbeds and toolsets
such as AutoGOLE / SENSE
Frederic Loui,
Casaba Mate,
Marcos Schwarz
et al.

18. SC15-22: SDN Next Generation
Terabit/sec Ecosystem for Exascale Science
SC16+: Consistent
Operations with
Agile Feedback
Major Science
Flow Classes
Up to High Water
Marks
supercomputing.caltech.edu
Tbps Rings for SC18-22: Caltech, Ciena, Scinet,
StarLight + Many HEP, Network, Vendor Partners
45
SDN-driven flow
steering, load
balancing, site
orchestration
Over Terabit/sec
Global Networks
LHC at SC15: Asynchronous Stageout
(ASO) with Caltech’s SDN Controller
Preview PetaByte
Transfers to/
from Site Edges of
Exascale Facilities
With
100G -1000G DTNs

19. Global Petascale to Exascale Workflows for Data Intensive
Sciences Accelerated by Next Generation Programmable
Network Architectures and Machine Learning Applications
 A Vast Partnership of Science and Computer Science Teams, R&E Networks
and R&D Projects
 Convened by the GNA-G Data Intensive Sciences WG
 Mission
 Meeting the challenges faced by leading edge data intensive experimental
programs in high energy physics, astrophysics, genomics and other fields
of data intensive science
 Clearing the path to the next round of discoveries
 Demonstrating a wide range of the latest advances in:
 Software defined and Terabit/sec networks
 Intelligent global operations and monitoring systems
 Workflow optimization methodologies with real time analytics
 State of the art long distance data transfer methods and tools,
local and metro optical networks and server designs
 Emerging technologies and concepts in programmable networks
and global-scale distributed system
Network Research Exhibition NRE-19. Abstract:
https://www.dropbox.com/s/qcm41g7f7etjxvy/SC22_NRE_GlobalPetascaleWorkflows_V8072022.docx?dl=0

20. Global Petascale to Exascale Workflows
Cornerstone system components and concepts (I)
 Integrated operations and orchestrated management of resources:
interworking with and advancing the site (Site-RM) and network resource
managers (Network-RM) developed in the SENSE program.
 An ontological model-driven framework with integration of an analytics engine,
API and workflow orchestrator extending work in the SENSE project, enhanced
by efficient multi-domain resource state abstractions and discovery mechanisms.
 Fine-grained end-to-end monitoring, data collection and analytics, supporting
applications with path selection + load balancing mechanisms optimized
by machine learning.
 The integration of Qualcomm Technology’s GradientGraph (G2) with
5G / OpenALTO / SRv6. These will be used to demonstrate how applications
including XR, auto/V2X, and IoT can benefit from the intelligent routing,
rate limiting, and service placement decisions computed by G2.
 Demonstrating how the integration of G2 with science networks, through the
integration of the LHC Rucio / FTS data management system, AutoGOLE / SENSE
virtual circuit and orchestration services / and OpenALTO monitoring may be used
towards optimizing large-scale data transfers, sharing data among CERN and
scientists at sites across the globe.

21. Global Petascale to Exascale Workflows
for Data Intensive Sciences
 Advances Embedded and Interoperate within a ‘composable’
architecture of subsystems, components and interfaces,
organized into several areas:
 Visibility: Monitoring and information tracking and management including
IETF ALTO/OpenALTO, BGP-LS, sFlow/NetFlow, Perfsonar, Traceroute,
Qualcomm Gradient Graph congestion information, Kubernetes statistics,
Prometheus, P4/Inband telemetry
 Intelligence: Stateful decisions using composable metrics (policy, priority,
network- and site-state, SLA constraints, responses to ‘events’ at sites and
in the networks, ...), using NetPredict, Hecate, RL-G2, Yale Bilevel
optimization, Coral, Elastiflow/Elastic Stack
 Controllability: SENSE/OpenNSA/AutoGOLE, P4/PINS, segment routing
with SRv6 and/or PolKA, BGP/PCEP
 Network OSes and Tools: GEANT RARE/freeRtr, SONIC, Calico VPP,
Bstruct-Mininet environment, ...
 Orchestration: SENSE, Kubernetes (+k8s namespace), dedicated code and
APIs for interoperation and progressive integration

22. Caltech, StarLight, NRL and Partners at SC22
Microcosm: Creating the Future of SCinet and of Networks for Science

23. GNA-G and its Working Groups
Worldwide Partnerships at SC22 and Beyond

24. Global Petascale to Exascale Workflows
Elements and Goals of the Demonstrations (Cont’d)
 400 Gbps Next Generation Networks
 High-Performance Routing of Science Network Traffic
 Creation of an overlay network with PolKA tunnels forming virtual circuits
 5G/Edge Computing Application Performance Optimization
 Integration of OpenALTO, SRv6 and Qualcomm Technologies’
GradientGraph
 N-DISE: Named Data Networking for Data Intensive Experiments
 Coral: Scalable Data Plane Checking via Distributed, On-Device Verification
 High performance networking with the Bella Link and the Sao Paulo
Backbone
 AmLight Express and Protect (AmLight-EXP)
 Global Research Platform (GRP)
 Software Defined International Open Exchanges (SDXs)
For further Information on NRE-19 Elements and Goals: See the Next Slides

25. NSF CC* Program: Integration and Innovation
SCIENTIFIC and BROADER IMPACT
▪ Lay groundwork for an NDN-based
data distribution and access system
for data-intensive science fields
▪ Benefit user community through
lowered costs, faster data access
and standardized naming structures
▪ Engage next generation of scientists
in emerging concepts of future
Internet architectures for
data intensive applications
▪ Advance, extend and test the NDN
paradigm to encompass the most
data intensive research applications
of global extent
SOLUTIONS + Deliverables
CHALLENGES
▪ LHC program in HEP is world’s largest
data intensive application: handling
One Exabyte by 2019 at hundreds
of sites
▪ Global data distribution, processing,
access, analysis; large but limited
computing, storage, network
resources
APPROACH
▪ Use Named Data Networking (NDN)
to redesign LHC HEP network;
optimize workflow
N-DISE: NDN for Data Intensive Science Experiments
TEAM
▪ Northeastern (PI: Edmund Yeh),
Caltech (PI: Harvey Newman) UCLA
(PI: Jason Cong), Tennessee Tech
(PI: Susmit Shannigrahi)
▪ In partnership with NIST, other LHC
sites and the NDN project team
▪ Simultaneously optimize caching
(“hot” datasets), forwarding, and
congestion control in both the
network core and site edges
▪ Development of naming scheme and
attributes for fast access and efficient
communication in HEP and other fields

26.  Feasibility of NDN-based data distribution system for LHC first
demonstrated at SC 18 (Dallas)
 SANDIE demo at SC 19 (Denver) showed improved throughput
and delay performance
 N-DISE demo at SC 21 (done remotely) showed further improvements, ICN
‘22 paper introduced the project to the literature
 SC22
 Jointly optimized caching and forwarding algorithms developed by
Northeastern
 High-speed NDN-DPDK forwarder developed by NIST
 NDN-based filesystem plugin for XRootD developed by Caltech
 FPGA acceleration of NDN-DPDK forwarder by UCLA
 Deploying genomics data lakes using K8s over NDN and name-based
compute placement to K8 clusters by Tennessee Tech
 Goals: (i) rapid and reliable delivery of LHC data over transcontinental
layer-2 demo testbed, (ii) demonstration of Kubernetes integration
for genomics applications
N-DISE

27. Caltech
Campus
A New Generation Persistent 400G/100G Super-DMZ: CENIC,
Pacific Wave, ESnet, Internet2, Caltech, UCSD, StarLight ++
SEA
100GE
SNVL
SDSC
LA
Pacific Wave
Seattle 1-1
Pacific Wave
Sunnyvale 2-1
NCS/Ciena
Transponders
LAX Agg10
Riser Panel
NCS/Ciena
Transponders
Riser Panel
Caltech IMSS
Waveserver Ai
200G
200G
Caltech IMSS
Waveserver Ai
CENIC LA
818 W 7th
6th Floor
CENIC LA
818 W 7th
10th Floor
Cisco NCS 1K4
Arista 7060 DX4
400GE & 100GE
Internet2
818 W 7th
10th Floor
Starlight
CENIC/
Pacific
Wave
I2 Optical
I2 Packet
To StarLight To SC22 Dallas
400G
400G
ESnet6 Fabric
ESnet/Fabric
818 W 7th
10th Floor
Pending Connections
to CENIC/PacWave
Pacific Wave
LosA 2-1
and Beyond
To Juniper QFX
after SC22

29. Next Generation Network-Integrated System
 Top Line Message: In order to meet the needs, we need to transition
to a new dynamic and adaptive software-driven system, which
 Coordinates worldwide networks as a first class resource
along with computing and storage, across multiple domains
 Simultaneously supports the LHC experiments, other major DIS programs
and the larger worldwide academic and research community
 Systems design approach: A global dynamic fabric that flexibly
allocates, balances and conserves the available network resources
Negotiating with site systems that aim to accelerate workflow; Use of ML
Builds on ongoing R&D projects: from regional caches/data lakes to
intelligent control and data planes to ML-based optimization
[E.g. SENSE/AutoGOLE, NOTED, ESNet HT, GEANT/RARE, AmLight, Fabric,
Bridges; NetPredict, DeepRoute, ALTO, PolKA ...]
 A key milestone: integration of SENSE + network services with FTS & Rucio
 We are also leveraging the worldwide move towards a fully programmable
ecosystem of networks and end-systems (P4, PINS; SRv6; PolKA),
plus operations platforms (OSG, NRP; global SENSE and P4 Testbeds)
 The GNA-G and its Working Groups, The LHC experiments together with the
WLCG and the worldwide R&E network community, are the key players
 Together with leaders and teams from: LBNF/DUNE, VRO, SKA

30. Global Petascale to Exascale Workflows
for Data Intensive Sciences
 Development Trajectory: Parallel developments + mission-driven
progressive interfacing and system-level integration
 Architecture: Data Center Analogue
 Classes of “Work” (work = transfers, or overall workflow),
defined by task parameters and/or priority and policy
 Adjust rate of progress in each class to respond to network
or site state changes, and “events”
 Moderate/balance the rates among the classes to optimize
a multivariate objective function with constraints
 Overarching Concept: Consistent Network Operations:
 Stable load balanced high throughput workflows cross optimally
chosen network paths
 Provided by autonomous site-resident services dynamically
interacting with network-resident services
 Up to preset or fliexible high water marks to accommodate
other traffic
 Responding to (or negotiating with) site demands from the science
programs’ principal data distribution and management systems

31. 31
Towards a Next Generation
Network-Integrated System
For Data Intensive Sciences
The Global Network
Advancement Group
The AutoGOLE/SENSE and
Data Intensive Sciences WGs
Partnering with GRP and NRP

32. Extra Slides Follow
32

33. Steps to Arrive at a Fully Functional System by 2028
the Data Challenge Perspective
 Three Types of Challenges
1. Functionality Challenge : Where we establish the functionality we
want in our software stack, and do so incrementally over time
2. Software Scalability Challenge: Where we take the products that
passed the previous challenge, and exercise them at full scale
but not on the final hardware infrastructure
3. End-to-end Systems Challenge: On the actual hardware; can only
be done once the actual hardware systems are in place.
 Targets are 2022-24 for 1 and 2 (or 2024-25 if not all
components are ready earlier); 2026-2027 for 3
 Remark: it’s conceivable, maybe even likely that it takes multiple
attempts to achieve sustained performance at scale with all of the
new software we need, with the functionality we want.
 + Scaling Challenges: Demonstrate capability to fill ~50% full bandwidth
required in the minimal scenario with production-like traffic: Storage to
storage, using third party copy protocols and data management services used
in production: 2021: 10%; 2023-4: 30%; 2025-6: 60%; 2028: 100%
33
F. Wuerthwein, UCSD/SDSC

34. Rednesp and RNP: Expanding Capacity
Among Latin America, US, Europe and Africa
 Rednesp (formerly ANSP) has 3 links to the
USA, connecting Sao Paulo to Florida.
 Atlantic 100G link: direct São Paulo- Miami
 Pacific 100G link: São Paulo to Santiago
(Chile),
from there to Panama, San Juan and Miami.
 200 Gbps link: São Paulo to Florida through
Fortaleza, on the Monet cable (Angola
cables).
 RNP also a 200 Gbps link using the Monet
cable:
 Total 600 Gbps capacity between Brazil and the
USA.
 Transatlantic Links: The Bella link between São
Paulo and Sines in Portugal, and a link
connecting São Paulo, Angola and South Africa.
 The deployment of the "Backbone SP" has just
started.
Its main goal is to interconnect most
universities in Sao Paulo (state) with 100 gbps
links.
• The first link connects SP4 to UNIFESP (Federal
University at São Paulo). Next, links should
connect UNESP (State University of São Paulo),
USP (University of São Paulo), UNICAMP

35. BRIDGES: “Binding Research Infrastructures for the Deployment
of Global Experimental Science” Across the Atlantic
 George Mason and East Carolina Universities (NSF IRNC Program)
 High Level View of the Infrastructure and Collaborators:
 Virtualization as an Architectural (systems-level) concept, targeting:
 Deterministic, predictable performance
 Agile, customizable, integrated virtual resource model
 Operates as a “fully virtualized” services environment: Users (and Operations personnel)
can interact with the BRIDGES services through an interactive web portal, or software agents
may interact via a programmatic API to automate resource provisioning and orchestration.

36. (APONet): Closing the Global Ring East and West
KREONet2: Closing the Global Ring East and West

37. BRIDGES: “Binding Research Infrastructures for the Deployment
of Global Experimental Science” Across the Atlantic

38. AER Update - Aug 2022
● Since the AER MoU, KAUST
is coordinating with REN
partners deployment of
sharing spare
capacity
● KAUST is supporting the
following partners by offering
point-to-point circuits for
submarine cable backup
paths:
○ AARnet
○ GÉANT
○ NetherLight
○ NII/SINET
○ SingAREN
● The SC22 NRE Demonstrations
will also be supported by KAUST
closing the ring from Amsterdam
to Singapore and back to the US
○ SC22 NRE
Seattle
Los
Angeles
Chicag
o
SC22 NRE Demos (Caltech)
KAUST and the AsiaPacific Oceana Network
(APONet): Closing the Global Ring East and West

39. 39
Gua
m
SingAR
EN
Hong
Kong
Daejeo
n Los
Angeles
Seattl
e
Chicag
o
Amsterda
m
Londo
n
Hawa
ii
Jedda
h
Toky
o
KAU
ST
Singapo
re
Cenic
KISTI
StarLigh
t
NII/SINE
T
NetherLig
ht
GÉANT
Open
Exchange
SC22 Caltech NRE Demonstrations
* The dotted paths in the diagram are the ones to be used for
the SC22
TransP
AC
KAU
ST
SingARE
N
NII /
SINET
KISTI
Univ. of
Hawaii
KAUST SC22 Demonstrations (NRE-19)

40. SC22 NREs: AutoGOLE / SENSE and NOTED (Draft)
Bill Johnston 9/23/22
SC21: Global footprint. Multiple 400G Optical Wide Area Links
SC22: Global footprint. Terabit/sec Triangle Starlight – McLean – Dallas; 400G to LA;
4 X 100G to Caltech and UCSD/SDSC; 3 X 400G to the Caltech Booth

41. Partner NREs at SC22
NRE-001
Edmund
Yeh
eyeh@ece.neu.edu N-DISE: NDN for Data Intensive Science Experiments
NRE-004
Joe
Mambretti
j-mambretti@northwestern.edu
1.2 Tbps Services WAN Services: Architecture,
Technology and Control Systems
NRE-005
Joe
Mambretti
j-mambretti@northwestern.edu
400 Gbps E2E WAN Services: Architecture, Technology
and Control Systems
NRE-007
Edoardo
Martelli
edoardo.martelli@cern.ch LHC Networking And NOTED
NRE-008
Joe
Mambretti
j-mambretti@northwestern.edu
IRNC Software Defined Exchange (SDX) Multi-Services
for Petascale Science
NRE-009 Jim Chen jim-chen@northwestern.edu
High Speed Network with International P4 Experimental
Networks for The Global Research Platform
and Other Research Platforms
NRE-010
Magnos
Martinello
magnos.martinello@ufes.br
Demonstrating PolKA Routing Approach to Support
Traffic Engineering for Data-intensive Science
NRE-011 Qiao Xiang xiangq27@gmail.com
Coral: Fast Data Plane Verification for Large-Scale
Science Networks via Distributed, On-Device Verification
NRE-013
Tom
Lehman
tlehman@es.net
AutoGOLE/SENSE: End-to-End Network Services
and Workflow Integration
NRE-015
Tom
Lehman
tlehman@es.net SENSE and Rucio/FTS/XRootD Interoperation
NRE-016
Marcos
Schwarz
marcos.schwarz@rnp.br
Programmable Networking with P4, GEANT RARE/freeRtr and
SONIC/PINS
The NRE demonstrations hosted at or partnering with the Caltech Booth 2820:

42. Global Petascale to Exascale Workflows
Cornerstone system components and concepts (II)
 Adapting NDN for data intensive sciences including advanced cache design and
algorithms and parallel code development and methods for fast and efficient access
over a global testbed, leveraging the experience in the NDN for Data Intensive Science
Experiments (N-DISE) project
 Demonstrating a paragon network at many sites equipped with P4 programmable
devices, including Tofino and Tofino2-based switches, Smart NICs and Xilinx FPGA-
based network interfaces providing packet-by- packet inspection, agile flow
management and state tracking, and real-time decisions
● High throughput platform demonstrations in support of workflows for the science
programs. Including reference designs of NVMe server systems to match a 400G
network core, as well as servers with multi-GPUs and programmable smart NICs
with FPGAs.
● Integration of edge-focused extreme telemetry data (from P4 switches, smart edge
devices and end hosts) and end facility/application caching stats and other metrics,
to facilitate automated decision-making processes.
● Development of dynamic regional caches or “data lakes” that treat nearby as
a unified data resource, building on the successful petabyte cache currently in
operation between Caltech and UCSD and in ESNet based on the XRootD federated
access protocol; extension of the cache concept to more remote sites such as
Fermilab, Nebraska and Vanderbilt.

43. Global Petascale to Exascale Workflows
Cornerstone system components and concepts (III)
● Creation of an overlay network with PolKA tunnels forming virtual circuits,
integrating persistent resources from the GNA-G AutoGOLE/SENSE and
GEANT RARE testbeds, to validate the PolKA source routing mechanisms
using 100G+ transfers of science data.
● Underlay congestion will be detected by tunnel monitoring and signaled
to the overlay so that the traffic is steered away from congested tunnels
to other paths.
● Comparisons between segment routing and PolKA regarding controllability and
performance metrics are also planned. From the application's point
of view.
● PolKA full deployment enables the extreme traffic engineering demands of data-
intensive sciences to be met, by offering a new range of network functionalities
such as: multipath routing, in-network telemetry and
proof-of-transit with path attributes to support higher level stateful traffic
engineering decisions.
● Network traffic prediction and engineering optimizations using the latest
graph neural network and other emerging deep learning methods, developed
by ESnet’s Hecate /DeepRoute project.

44.  LHC:
 End to end workflows for large scale data distribution and analysis in support of the CMS
experiment’s LHC workflow among Caltech, UCSD, LBL + Fermilab, Nebraska, Vanderbilt,
and GridUNESP (Sao Paulo) including automated flow steering, negotiation and DTN
autoconfiguration; bursting of some of these workflows to the NERSC HPC facility and the
cloud; use of both edge and in-network caches to increase data access and processing
efficiency.
 An ontological model-driven framework with integration of an analytics engine, API and
workflow orchestrator extending work in the SENSE project, enhanced by efficient multi-
domain resource state abstractions and discovery mechanisms.
 SENSE/AutoGOLE:
 The GNA-G SENSE/AutoGOLE WG demonstration will present key technologies, methods
and a system of dynamic Layer 2 and Layer 3 virtual circuit services to meet the challenges
and address the requirements of the largest data intensive science programs, including the
Large Hadron Collider (LHC) the Vera Rubin Observatory and programs challenges and
programs in many other disciplines. The services are designed to support multiple petabyte
transactions across a global footprint, represented by a persistent testbed spanning the US,
Europe, Asia Pacific and Latin American regions.
 Global Ring and KAUST: These demonstrations will also showcase the power of
collaboration in the global research and education network (REN) community. The demo will
have KAUST as a collaborator in the Asia-Pacific Global Ring (AER [*]), closing the global ring
by interconnecting Amsterdam to Singapore, and then onto Los Angeles with support from
partners including SingaREN, JGN-X and APOnet.
 [*] See Fast and stable Network Ring between Asia-Pacific and Europe | AARNet
https://www.aarnet.edu.au/fast-and-stable-network-ring-between-asia-pacific-and-europe/
Global Petascale to Exascale Workflows
Elements and Goals of the Demonstrations (I)

45. Global Petascale to Exascale Workflows
Elements and Goals of the Demonstrations (II)
 400 Gbps Next Generation Networks
 Science research collaborations continuing to prepare for increased network capacity, e.g.,
transitioning from 100 Gbps paths to 400 Gbps, 800 Gbps and beyond. An international
collaboration led by the StarLight/ Caltech/ UCSD/ NRP/GRP/SCinet consortium, with many
will demonstrate the architecture, services and techniques to enable efficient management of
high volume science data streams within multi-hundred Gbps channels.
 Global Research Platform (GRP):
 An international collaboration will demonstrate services and capabilities of the prototype
Global Research Platform, an international Science DMZ, optimized for large scale world-wide
science projects, especially those based on instruments producing high volumes of data.
 Software Defined Exchanges (SDXs):
 Several international open exchanges that are developing capabilities for Software Defined
Exchanges (SDXs) will showcase the capabilities of these facilities for providing highly
programmable capabilities for international science data transport.
 AmLight Express and Protect (AmLight-EXP)
 In support of the SENSE/AutoGOLE demonstrations and LHC-related use cases will be
shown, in association with high-throughput low latency experiments, and demonstrations of
auto-recovery from network events, using optical spectrum on the Monet submarine cable,
and its 100G ring network that interconnects the research and education communities in the
U.S. and South America.

46. Global Petascale to Exascale Workflows
Elements and Goals of the Demonstrations (III)
 5G/Edge Computing Application Performance Optimization
 UCSD, the Pacific Research Platform, Caltech, Yale and Qualcomm Technologies will attempt to demonstrate
the first end-to-end integrated traffic optimization framework based on bottleneck structure analysis for 5G
applications. This intended demonstration will focus on showing the integration of Qualcomm technology’s
GradientGraph with the IETF ALTO open standard, to support the optimization of edge computing
applications such as XR, holographic telepresence, and vehicle networks. Holodecks at UCSD and NYU will
be interconnected across the CENIC and NYSERNet regional networks via a transcontinental link.
 High-Performance Routing of Science Network Traffic
 UCSD, the Pacific Research Platform, Caltech, Yale and Qualcomm Technologies will attempt to demonstrate
the first end-to-end integrated traffic optimization framework based on bottleneck structure analysis for 5G
applications. This intended demonstration will focus on showing the integration of Qualcomm technology’s
GradientGraph with the IETF ALTO open standard, to support the optimization of edge computing
applications such as XR, holographic telepresence, and vehicle networks. Holodecks at UCSD and NYU
will be interconnected across the CENIC and NYSERNet regional networks via a transcontinental link.
 Integration of OpenALTO, SRv6 and Qualcomm Technologies’ GradientGraph
 We intend to demonstrate the integration of OpenALTO, SRv6 and Qualcomm Technologies’ GradientGraph,
showing how applications including XR, Auto/V2X, and holographic telepresence can benefit from the
intelligent routing, rate limiting, and service placement decisions computed by the GradientGraph platform.
We also intend to demonstrate the integration of OpenALTO and GradientGraph with science networks,
Rucio, FTS, AutoGOLE, SENSE and OpenALTO towards optimizing large-scale data transfers from the
Large Hadron Collider at CERN to scientists across the globe.

47. Global Petascale to Exascale Workflows
Elements and Goals of the Demonstrations (IV)
 SENSE: The Software-defined network for End-to-end Networked Science at Exascale
 The SENSE project is building smart network services to accelerate scientific discovery in the
era of ‘big data’ driven by Exascale, cloud computing, machine learning and AI. The SENSE
SC22 demonstration showcases a comprehensive approach to request and provision end- to-
end network services across domains that combines deployment of infrastructure across
multiple labs/campuses, SC booths and WAN with a focus on usability, performance and
resilience through:
 Priority QoS for SENSE enabled flows
 Multi-point and point-to-point services
 Auto-provisioning of network devices and Data Transfer Nodes (DTNs);
 Intent-based, interactive, real time application interfaces providing intuitive access
to intelligent SDN services for Virtual Organization (VO) services and managers;
 Policy-guided end-to-end orchestration of network resources, coordinated with the
science programs' systems, to enable real time orchestration of computing and
storage resources.
 Real time network measurement, analytics and feedback to provide the foundation
for full lifecycle status, problem resolution, resilience and coordination between
the SENSE intelligent network services, and the science programs' system services.

48. Global Petascale to Exascale Workflows
Elements and Goals of the Demonstrations (V)
 N-DISE: Named Data Networking for Data Intensive Science Experiments
 The NDN for Data Intensive Science Experiments (N-DISE) project aims to accelerate the pace of
breakthroughs and innovations in data-intensive science fields such as the Large Hadron Collider
(LHC) high energy physics program and the BioGenome and human genome projects. Based on
Named Data Networking (NDN), a data-centric future Internet architecture, N-DISE will deploy and
commission a highly efficient and field-tested petascale data distribution, caching, access and
analysis system serving major science programs. The N-DISE project will build on recently
developed high-throughput NDN caching and forwarding methods, containerization techniques,
leverage the integration of NDN and SDN systems concepts and algorithms with the mainstream
data distribution, processing, and management systems of CMS, as well as the integration with
Field Programmable Gate Arrays (FPGA) acceleration subsystems, to produce a system capable
of delivering LHC and genomic data over a wide area network at throughputs approaching 100
Gbits per second, while dramatically decreasing download times. N-DISE will leverage existing
infrastructure and build an enhanced testbed with high performance NDN data cache servers at
participating institutions.
 The N-DISE demonstration is designed to exhibit improved performance of the N-DISE system for
workflow acceleration within large-scale data-intensive programs such as the LHC high energy
physics, BioGenome and human genome programs. To achieve high performance, the
demonstration will leverage the following key components: (1) the transparent integration of NDN
with the current CMS software stack via an NDN based XRootD Open Storage System plugin, (2)
joint caching and multipath forwarding capabilities of the VIP algorithm, (3) integration with FPGA
acceleration subsystems, (4) SDN support for NDN through the work of the Global Network
Advancement Group (GNA-G) and its AutoGOLE/SENSE and Data Intensive Sciences Working
Group.

49. Global Petascale to Exascale Workflows
Elements and Goals of the Demonstrations (VI)
 High performance networking with the Bella Link and the Sao Paulo Backbone
 With the recent availability of an L2 circuit linking the University of São Paulo State
(UNESP) to CERN in Geneva, through the Bella Link, Brazil (and São Paulo State) now has
three 100 gbps academic links to the USA and to Europe. This new circuit is supposed to
present low latency, and that should open new opportunities for collaboration among
Brazilian and European academic institutions. All links pass through the SP4 Equinix
datacenter in the greater Sao Paulo area. At the SP4 datacenter, rednesp (Research and
Education Network at Sao Paulo) has a node called SouthernLight which is part of the GNA
Autogole/SENSE testbed.
 At the same time, the Rednesp Sao Paulo regional network in finishing the "Backbone SP"
connecting 9 important universities in Sao Paulo state. In this demo, we intend to compare
properties such as effective bandwidth, latency to different continents and jitter using
different paths of the 3 intercontinental links as well as the integration of the "Backbone
SP" to international connections. We also expect to show results and metrics of parallel
computations using some of the national and international computing facilities connected
to this infrastructure.

50. Global Petascale to Exascale Workflows
Elements and Goals of the Demonstrations (VII)
 Coral: Scalable Data Plane Checking via Distributed, On-Device Verification
 Data plane verification (DPV) is important for finding network errors, and therefore a fundamental
pillar for achieving consistently operating, autonomous-driving, high-performance science
networks. Current DPV tools employ a centralized architecture, where a server collects the data
planes of all devices and verifies them. Despite substantial efforts on accelerating DPV, this
centralized architecture is inherently non-scalable. To tackle the scalability challenge of DPV, the
team of Xiamen University, China, circumvents the scalability bottleneck of centralized design
and design Coral, a distributed, on-device DPV framework.
 The key insight of Coral is that DPV can be transformed into a counting problem on a directed
acyclic graph, which can be naturally decomposed into lightweight tasks executed at network
devices, enabling scalability. Coral consists of (1) a declarative requirement specification
language, (2) a planner that employs a novel data structure DVNet to systematically decompose
global verification into on-device counting tasks, and (3) a distributed verification (DV) protocol
that specifies how on-device verifiers communicate task results efficiently to collaboratively
verify the requirements. During SC'22, we will demonstrate, through both a testbed
reconstruction of the Internet2 WAN environment and large-scale simulations of WAN/LAN/DC
with real-world datasets, that Coral consistently achieves scalable DPV under various scenarios.
A preliminary manuscript of Coral can be found at [1] and a set of small-scale demos can be
found at [2].
 [1] Qiao Xiang et al., Switch as a Verifier: Toward Scalable Data Plane Checking via Distributed, On-
Device Verification, https://arxiv.org/abs/2205.07808
 [2] Qiao Xiang et al., Coral System Functionality Demonstration, http://distributeddpvdemo.tech/

51. Network Traffic Evolution 2021-22 to and from CERN
LHCOPN LHCOPN Traffic CERN to Tier1s
Overall CERN Traffic: LHCOPN, LHCONE, Internet
Out
In
E. Martelli, CERN/IT

52. HL-LHC Network Needs and Data Challenges
Current Understanding: Q2 2021
 Export of Raw Data from CERN to the Tier1s (350 Pbytes/Year):
 400 Gbps Flat each for ATLAS and CMS Tier1s;
+100G each for other data formats; +100 G each for ALICE, LHCb
 “Minimal” Scenario [*]: Network Infrastructure from CERN to Tier1s Required
 4.8 Tbps Aggregate: Includes 1.2 Tbps Flat (24 X 7 X 365) from the above,
x2 to Accommodate Bursts, and x2 for overprovisioning, for operational
headroom: including both non-LHC use, and other LHC use.
 This includes 1.4 Tbps Across the Atlantic for ATLAS and CMS alone
 Note that the above Minimal scenario is where the network is treated as a
scarce resource, unlike LHC Run1 and Run2 experience in 2009-18.
 In a “Flexible Scenario” [**]: 9.6 Tbps, including 2.7 Tbps Across the Atlantic
Leveraging the Network to obtain more flexibility in workload scheduling,
increase efficiency, improve turnaround time for production & analysis
 In this scenario: Links to Larger Tier1s in the US and Europe: ~ 1 Tbps
(some more); Links to Other Tier1s: ~500 Gbps
 Tier2 provisioning: 400Gbps bursts, 100G Yearly Avg: ~Petabyte Import in a shift
 Need to work with campuses to accommodate this: it may take years
[*] NOTE: Matches numbers presented at ESnet Requirements Review (Summer 2020)
[**] NOTE: Matches numbers presented at the January 2020 LHCONE/LHCOPN Meeting

53. (Southern) California ((So)Cal) Cache
53
Roughly 30,000 cores across Caltech & UCSD … half typically used for analysis
A 2 Pbyte Working Example in Production
Plan to include Riverside and other SoCal Tier3s; ESnet plan to install additional in-
network caches near US Tier2s; Move to Ceph underway
Scaling to HL LHC: ~30 Pbytes Per Tier2, ~10 Pbyte Caches, 1 Pbyte Refresh in a Shift
Requires 400G Link. Still relies on use of compact event forms,
efficiently managed data transport

54. SENSE: SDN for End-to-end Networked
Science at the Exascale
ESnet Caltech Fermilab LBNL/NERSC Argonne Maryland
54
 Use Cases: Inter-DTN Priority Flows, LHC File Transfer Service (FTS),
SENSE/AutoGole Integration (GNA-G), ExaFEL, Big Data Express

55. Internet2
AL2S
ESnet
AutoGOLE / SENSE Deployment
StarLight
NetherLight SURFNet
MOXY
CENIC
PacWave
RNP
AMPATH
Southern
Light
CERN
UCSD
LOSA
SEAT
SUNY
KREONET/
KRLight
KISTI
AMPATH
RNP
MANLAN
WIX
Site/End Systems Exchange Point Network
ESnet
Testbed
STAR

56. SENSE: SDN for End-to-end Networked
Science at the Exascale
ESnet Caltech Fermilab LBNL/NERSC Argonne Maryland
56

57. SENSE: SDN for End-to-end Networked
Science at the Exascale
ESnet Caltech Fermilab LBNL/NERSC Argonne Maryland
57

58. Self-driving Networks| BERKELEY LAB
Self-Driving Networks
“Networks should learn to drive themselves”*
58
[*] Why (and How) Networks Should Run Themselves,
Feamster and Rexford
https://doi.org/10.48550/arXiv.1710.11583
Example controllers:
Ryu, OpenDaylight,
OSCARS
Listener APIs
Supervised/Unsupervised
classification:
anomaly detection
Regression: forecasting
Action
plan
Network
Telemetry
Execute
Control
Reward function Inputs to AI agent
Try different
optimization
strategies and/or
objective functions
Can take simple actions to achieve key goals. Such as:
improving availability, attack resilience and dealing with scale.
Our argument is AI is needed for mission critical actions.

59. M. Kiran, C, Guok Talk at APAN2021
Intelligent Controller (HECATE)
Case Studies:
1. Model free: Path selection for large data transfers: better load balancing
2. Model Free: Forwarding decisions for complex network topologies:
Deep RL to learn optimal packet delivery policies vs. network load level
3. Model Based: Predicting network patterns with Netpredict
Self Driving Network
Adaptive Routing (e.g. Real
time data for routing decisions)
Learns to Avoid Congestion
Congestion Free  Loss Free
Towards 100% Utilization
Proactive Fault Repair
Mariam Kiran (ESnet) et al. Intelligent Networks DOE Project
Use Deep Reinforcement
Learning to Optimize network
traffic engineering

60. Self-Driving Network for Science
60
Model-based
learning
Model-free
learning
Predicting Future
Congestion
Study traffic
patterns
action
observations
e.g.Network
telemetry or
monitoring
reward e.g.
flow
completion
time
e.g.
Reconfigure
network e.g.
update flow
rules
Using Deep Reinforcement Learning for
optimizing traffic engineering over
networks
Kiran et al. Intelligent Networks DOE Project
 D-DCRNN: Spatiotemporal forecasting with applications in neuroscience, climate,
traffic flow, smart grid, logistics, supply chain: https://arxiv.org/pdf/1707.01926.pdf
 Future: Hooks for a Richer, More Stateful Objective Function
(Policy, Priority, Deadlines, Path Quality, Co-Flows...);
Event Response; Applications to 5G and Quantum Networks
Mariam Kiran (ESnet)
mkiran@lbl.gov et al

61. Self-driving Networks| BERKELEY LAB
HECATE Simulation
61
UR: link utilization rate
Moves incoming traffic
to less used paths

62. Science Networks are Expanding to Wireless 5G, 6G
Sensor Networks and Beyond
62
 Statically deployed, unattended network
sensors need full-time monitoring
 While sensors in mobile networks
(including unmanned aerial systems)
can self-manage and optimize data
collection
 For continual monitoring of the sites, data
is processed in real-time and transmitted
to a central location with 5G speed for
further processing
 Novel mesh network topologies for flexible
inclusion of new sources - much like new
cells in the body - are being developed
 Can use 5G latency to handle the tradeoff
between edge and cloud processing

63. SENSE: SDN Enabled Networks for Science at the Exascale
ESnet, Caltech, Fermilab, LBNL https://arxiv.org/abs/2004.05953
Creates Virtual Circuit Overlays. Site and Network RMs, Orchestrator

64. SC20 to 22+: AutoGOLE/SENSE Persistent Testbed:
ESnet, SURFnet, Internet2, StarLight, CENIC, Pacific Wave, AmLight, RNP, USP, UFES, GridUNESP,
Hawaii, Guam, KISTI, Tokyo,Caltech, UCSD, PRP, FIU, CERN, Fermilab, UMd, DE-KIT
Federation
with the StarLight
GEANT/RARE
and AmLight
P4 Testbeds
Underway
2022-23 Outlook:
ESnet6/High Touch
KAUST
FABRIC
BRIDGES
US CMS Tier2s
GridUNESP, UERJ
SKAO
AarNet
TIFR et al
APONet
Courtesy T. Lehman
Caltech/
UCSD/
LA 100G to 4 X
100 to 400G
with CENIC
Enhanced for
SC22
and Beyond
400G Link(s)
NetherLight-
CERN
Automation
Following
SENSE, NRP,
ALTO, Atlantic
Wave SDX
Persistent Operations: Beginning this Year
NRP

65. ●Goal is to more flexibly control
how these are utilized on
a per flow, group, or use basis
●Do not want to manage "every"
flow in the network, however
should be able to manage
"any" flow in the network
●There are multiple transatlantic
and transpacific links, operated
by multiple organizations
transatlantic links
Important Link Management
● Two timescales: SENSE overlay network of virtual circuits
with BW guarantees is relatively stable; IPv6 subnets and
Directors provide more dynamic flow handling
● An equally important
goal is to understand
the load and leave room
for other traffic
● Compatible with other
network operations
● Result is to make better
use of the available
resources overall

66. SENSE Services for LHC and Other Open Science Grid
(OSG) Workflows: Rucio, FTS, XRootD Integration
 Goals center around SENSE providing key network-related
capabilities to LHC + 30 other science programs
 Primary Motivations:
 Developing mechanisms for an application workflow to obtain information
regarding the network services, capabilities, and options;
to a degree similar to what is possible for compute resources
 Giving applications the ability to interact with the network:
to exchange information, negotiate performance parameters,
discover expected performance metrics, and receive status/
troubleshooting information in real time, during long transactions.
 Key pathways:
(1) Interfacing and interaction with RUCIO, FTS and XRootD data federation
and storage systems, through Engagement:
with CMS, ATLAS and those development teams
(2) Enabled by the ESnet6 plan: with up to ~75% of link capacity
on key routes available for policy-driven dynamic provisioning
and management by ~2027-28
 4 Phase 1 Year Schedule, Completing in Q1 2023:
from system evaluation to design, to prototype deployments and tests,
to a plan for the transition to operations and the additional R&D needed 66

67. European Science
Data Center
OSG Data Federation
Vera Rubin Observatory
Interfacing to Multiple VOs With FTS/Rucio/XRootD
LHC, Dark Matter, n, Heavy Ions, VRO, SKAO, LIGO/Virgo/Kagra; Bioinformatics

68. Beyond Programmability Alone: A Systems Approach
Qualcomm Technologies GradientGraph Analytics
68
 System Wide
information to identify,
deal with root causes
Key Components
• Bottleneck precedence
+ flow gradient graphs
• Impactful flow and
flow group ID
• Alternate path
recommendations
www.reservoir.com/gradientgraph/
 Objective: Flow performance optimization in high
speed networks, with fairness
 Approach: Built on a new mathematical Theory of
Bottleneck Structures and an analytical framework
 Enables operators to understand and precisely control
flow(-group) and bottleneck performance
 Value: Improved capacity planning, traffic engineering
 Greater, more effective network throughput and stability
as a function of capacity and cost
 "On the Bottleneck Structure of Congestion-Controlled Networks," ACM
SIGMETRICS, Boston, June 2020 [https://bit.ly/3eGOPrb]
 "Designing Data Center Networks Using Bottleneck Structures,"
accepted for publication at ACM SIGCOMM 2021 [https://bit.ly/2UZCb1M]
 "Computing Bottleneck Structures at Scale for High-Precision Network
Performance Analysis," SC 2020 INDIS, November 2020
[https://bit.ly/3BriwaB]
 "A Quantitative Theory of Bottleneck Structures for Data Networks",
Technical Report, available upon request [https://bit.ly/38u8ARs]

69. Global P4 Lab and Production Grade
Programmable Switches
Marcos Schwarz
SC22
November 14-17 2022, Dallas - TX

70. Motivation
Can we increase the rate of evolution without interfering with production networks?
To develop and operate end-to-end / multi-domain orchestration services:
• Resource reservation (guaranteed bandwidth)
• Resource provisioning (Circuits, VRFs)
• Underlay observability
• Dynamic traffic steering/engineering
• Dynamic creation of L3 VPNs
• Closed loop multi-domain visibility/intelligence/controllability
How can we create/sustain an integration initiative/platform to propose and
validate next generation protocol and services?
• Proposition: Use programmable P4 devices to experiment on pre-production
networks leveraging industry/R&E open ecosystems

71. Global P4 Lab Goals
Current (SC22)
• Persistent global L3 overlay network based on P4 switches
• Intercontinental high capacity transfers (100G and over) exploring multiple
source routing solutions (Segment Routing and PolKA)
• Management infrastructure and tools used to operate this global network
• Core network based on RARE/FreeRtr and edge networks based also on SONiC
Future (2023)
• Capability to support multiple virtual networks, that implement on the same
devices different choices of routing stacks, traditional and SDN based
• Integration with initiatives for visibility, controllability and intelligence
• Work as a reference state of the art / next generation R&E network

72. Integration Path with other Initiatives