Montana State University and the First National Research Platform Workshop
Jerry Sheehan, Vice President for Information Technology
and Chief Information Officer for Montana State University
Tae Hwang, Technical Solutions Architect, Cisco
November 3, 2017 #msubridger
Overview of Presentation
I. Why Montana State Built a ScienceDMZ and the Practical Reasons that Higher Education CIOs Should Care about Research Networking
II. The Pacific Research Platform and the 1st Workshop for the National Research Platform
III. Why Montana State is Interested in Cisco SDA and Software DNA
IV. The Cisco Approach to Research Networking
Montana State, A LEAN CI Approach
• Campus is 16,440 students, 1,302 faculty and 2,016 staff
• Core Information Technology Budget is ~$14M with 75 FTE
• Research Cyberinfrastructure Staff: 2 FTE, positions a combination of technical and outreach • RCI supports: Community HPC cluster, Collaboration environments, and the
Bridger research network
• Networking Staff is 3 FTE for 35,000 (20K wireless, 15K wired) devices
Technology Adoption Curves and Research Networking
Image from: Hoonuit.com, Understanding the Technology Curve in Education,blog.atmoiclearning.com
The Day Job of a Higher Education Chief Information Officer
• The Primary Decision that CIOs make are Economic• Limited resources (financial and personnel) and unlimited demand (commoditized
technology/user needs and wants)
• The Smaller the Size of the Campus the More Sensitive Each Resource Allocation Decision Becomes
• Research Intensive Campuses can be Early Adopters if there is a Demand for Services
• Research Networking Suffers Across All Campuses Because Most Scientists Don’t Know What’s Possible Possible
Why Montana State Created our ScienceDMZ
100GBFile
Montana State’s Research Networking Infrastructure
Science Outreach Via Film {Dennis Aig}
Climate Modeling {Ben Poulter}
Virology {Mark Young}
Marmoset Vocalization {Ross Snider}
Optical Remote Sensing {Joe Shaw}
Digital Microscopy {Phil Stewart}
Carbon Sequestration {Lee Spengler}
Vegetation Modeling {Dave Roberts}
GIS {Rick Lawrence}
Montana State Science Drivers
Research Networking Has a History of HavingMajors Impact on Higher Education
A visualization study of NSFNET backbone traffic in September 1991
The traffic volume ranges from zero bytes (purple) to 100 billion bytes (white). It represents data collected by Merit Network, Inc.
In 1985, with the creation of the supercomputer centers program, NSF created NSFNET, a network that connected the five supercomputer centers and provided a network for research and education
Based on the ARPANET protocols, the NSFNET created a national backbone service
Visualization by Donna Cox and Robert Patterson, Courtesy of the National Center for Supercomputing Applications
Today’s Research Networks as the Bird Dogs for Tomorrow’s Enterprise Networks
• Software Defined Networks: Research Driving Usage, and with Private Sector Data Centers, Driving Vendors to “Bake It In”
• Persistent Network Monitoring with Tools Like GridFTP, MadDASH for Regions, and Probes Like PerfSonar
• Globus for Dealing with Identity, Graceful Large File Transfer, and Monitoring of How the Network is Moving Data
• Changing the Conversation with the End User Through Hybrid Staff like the CI-Facilitator (programs like ACIREF and CaRC) to Make it Clear the Network is a Dynamic Tool that Can Be Used to Empower Work.
• Building Interoperable Regional Networks for Persistent Big Data Transfer (Pacific Research Platform)
National Science Foundation Campus Cyberinfrastructure Investments Over the Last Half Decade
(GDC)
The Pacific Research Platform Creates a Regional End-to-End Science-Driven “Big Data Superhighway” System
NSF CC*DNI Grant$5M 10/2015-10/2020
PI: Larry Smarr, UC San Diego Calit2Co-Pis:• Camille Crittenden, UC Berkeley CITRIS, • Tom DeFanti, UC San Diego Calit2, • Philip Papadopoulos, UCSD SDSC, • Frank Wuerthwein, UCSD Physics and SDSC
Letters of Commitment from:• 50 Researchers from 15 Campuses• 32 IT/Network Organization Leaders
The First National Research Platform Workshop
Co-Chairs: Larry Smarr (Calit2) and Jim Bottum (Internet2)
Steering Committee: Alex Feltus (Clemson), Tracey Futhey (Duke), Ron Hutchins (University of Virginia), John Moore, (Internet2), Maria Meehl (UCAR), Jerry Sheehan (Montana State), Camile Crittenden (Berkeley), Tom DeFanti (UC San Diego), Amy Phillips (PNWG), Steve Riter (UTEP), Joe Mambretti (Northwestern), Inder Monga (EsNet), Jen Leasure(Quilt), Mount Allen (SF Jazz), Julio Ibarra (Florida International), Wendy Huntoon (KINBER), Ana Hunsinger(Internet2), Mark Johnson (MCNC), Richard Alo (Jackson State), Jason Arviso (Navajo Technical), Greg Monaco (Great Plains Network), Steve Hunter (Oregon), Maxine Brown (UIC), Louis Fox (CENIC)
http://prp.ucsd.edu/presentations/nrp
Findings from the First National Research Platform WorkshopUser Requirements1. NRP Needs to Be Easy For Scientists to Implement and Use2. Scientists Want to do Science, Not Networking or IT7. Voluntary Institutional and Researcher Commitment is Vital for Success11. Trust is a Key Element of Success
End User Engagement3. This is a Social Engineering Projects as Well as a Technical Networking IT
Project5. The Science Engagement Process is Crucial to Scaling Up at a
National/International Research Platform12. Need to think About Science Engagement Career Path and How to
Attract and Retain Good Staff
Technical4. The Science DMZ/DTN Architecture is an Effective Means for Enabling
High Performance Networking for Campuses, Balancing Researcher Requirements with Network Security Concerns
6. Several Classes of Cyberinfrastructure Will Benefit Significantly and Synergistically by Being Adapted to the PRP Model (National HPC, Instruments and Facilities, Data Repositories and Portals, Campus Computing Centers, Commercial Cloud Services)
8. NRP Needs to Be More than Interconnected DMZs9. The NRP Should Be Prepared to Evolve at the Same Time it is Being
Launched10. Tools and Protocols for Scaling Up Should be Developed and Shared13. Improved Measurement Techniques Should be Developed and Adopted14. Security Arose as a Thread Only Partially Addressed by Trust
Recommendations from the First NRP Workshop• The NSF should not pursue just incremental changes over the next 3-5 years, but rather transformative
changes. U.S. campuses cyberinfrastructure capabilities have advanced greatly under NSF’s Campus Cyberinfrastructure (CC*) programs, laying the foundation for an emerging NRP.
• Specifically, we recommend that NSF:• Continue funding its Campus Cyberinfrastructure program, including more opportunities for campus
cyber-engineers and cyber teams.• Issue a call for proposals to address the tough technical and sociopolitical issues that will need to be
addressed in an NRP; in fact, this could be a excellent opportunity for DOE and NSF to work together on a joint program.
• Engage science engagement facilitators on campuses to collaborate with each other across campuses in support of building the NRP.
• Support more regional Science DMZs to be formed from existing CC* grants to campuses. Multi-campus networking organizations should be encouraged to take the initiative to create regional DMZ proposals, including campuses that have not previously received NSF CC* grants. Parallelism is the key to scaling rather than central control.
Why is Montana State Excited About Cisco’s Software Define Access and the DNA Center?
• Simplifies the Operation of a Research Network, No Longer Separate and Equal, But Instead Converged.• Less hardware and leveraging the hardware you deploy to do
more.
• Increased Security• The Science DMZ says let’s harden by DTN because we aren’t
putting it behind the firewall (in/out), SDA offers the opportunity to intelligently route and continue to use the firewall.
• Increased Flexibility• Bridger deployment is dependent on location in buildings
because of required hardware. SDA frees us from having the hardware in the closet before we can help a researcher.
• The Promise of Plug and Play• Letting routing hardware auto-discover, turning hours of
configuration into moments of deployment.
• Unified and Easier Monitoring• Using DNA, see deeper and better into your network with one
platform.
NSF CC* and Campus IT
CampusIT
RegionalEducationNetworkOn/Off-Prem Computing/Storage
ScienceDMZNetwork FederationPerfSonar
DataTransferNodeAudit
FastAccess
ProvidesUnhinderedResearchCollaboration
HidesITComplexity
BTW,NSFCC*=ScienceDMZ?
Cisco’s Approach
ReferenceArchitecture
OurPeopleasTrustedAdvisorDesignand*ProposalAssistance*
Solution/Integration
Campus/WAN/I2NetworkingHighPerformanceComputingMachineLearning/AI,Storage
Security-VisibilityIntotrafficbothONandOFFNetwork
NSF CC* Customer Solution Example
GlobusDataTransferNodeFlashDrives10TB.Providesresearchcollaborations
HPCFlashParallelFileStorageActsaslocalscratch
ExistingIntelHPCCluster
HPCClusterwith(3)NvidiaP100GPUswithfutureexpansionwithexternalgpufarms(9GPUs/farm)
HPCClustermanagementandMachineLearningmgmt.BrightwillmanageHPCburstintoclouds.
Objectstorageforallofresearchers’data
NFSMounted(and/orCIFS)
NFS/SCP/FTPor/andGlobus
CLI/API/Client
I2/InternetBurst
25/10/5Tflops (FP16/32/64)
SecurityConsole
ASR1002
Nexus9200sSciecneDMZ switches
Perfsonar NetworkMonitoring- UCS220OnCiscoSwitches*
