Wat verandert het toekomstige Internet voor architecten? Sogeti DYA Dag 2017 Leon Gommans Science Officer Air France KLM Group IT Technology Office R&D Guest Researcher, University of Amsterdam FNWI- SNE group.
Content Introductie Probleem: large volume data transfer Concepten: Lightpath e- Infrastructure Internet Slicing Data Transfer Node Implicaties voor architectuurdenken Infra level: Big Data Sharing architectures Business level: Digital Market Place
The Internet and me RFC 2903-06 Research my first email ADP Network Services Enterprise LANs TCP/IP in LANs PhD 2014
Researching Future Internet capabilities? 1: Virtually infinite (dedicated) bandwidth across the globe Can data and processing be location independent? 2: A software definable infrastructure with virtualized functions. Ownership & Trust models for data & processing? 3: An infrastructure as a digital market place Market members share algorithms & data whilst ensuring mutual benefits?
Why KLM? Data in Aircraft Maintenance, Repair & Overhaul Fleet 20.000 Total Million of Terabytes (Exabytes) generated / year 100 By 2026 overall fleet of nearly 20.00 aircraft will generate 98 Exabytes/year 10.000 50 0 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 0 Source: Oliver Wyman Fleet & MRO Forecast, www.planestats.com/betterinsight
2000: Big Data at CERN ATLAS detector at CERN 7000 ton 46m x 25m x 25m generates 1 Pb/s of raw data. 1 of 4 detectors After filtering known events still leaves 10 s of PBs/year Problem: How to distribute such data volumes across the globe without being a DoS attack to the Internet
Netherlight: Europe s next generation Open Exchange www.netherlight.net
Global Lightpath e- Infrastructure www.glif.is Optical Network Exchanges
2004 Remote Visualisation across lightpaths Source:www.sagecommons.org Electronic Visualisation Lab at University of Illinois Chicago
2008 NSF GENI project: Internet Slicing www.geni.net University of Amsterdam SNE OpenLab Air France KLM Fieldlab GENI provides compute & network resources that can be deployed in a software defined way. GENI Rack
Internet Slicing: Future Logical view Vision: The Internet as a programmable virtual infrastructure with secure slices delivered by multiple autonomous service provider and enterprises My application runs in my slice containing resources from multiple datacenters and networks. KL Rack #1 Research Testbed Backbone #1 Backbone #2 AFI Rack #2 Mobile Access AF Rack #3 My Internet Slice Boeing / Airbus / GE Commercial cloud Digital Airport Infrastructure
Problem: Large data transfer across distance Gb/s 10 Local Metro 5 Regional Throughput with 0.005% packet loss Continental Inter- Continental 0 0 50 100 Distance (ms)
2013 ESNet Data Transfer Node technology Source: fasterdata.es.net Part of US Department of Energy
2015 NSF Pacific Research Platform Testbed Source: prp.ucsd.edu ExoGENI Testbed
Ambition: AF/KL Future Internet Field Lab
Researching Big Data Sharing: Use Cases Global Scale Aircraft MRO NWO/STW CIMPLO project Cybersecurity NWO COMMIT/ SARNET project National Scale Logistics Data sharing NLIP ishare project City / regional Scale Passenger Flows Campus / Enterprise Scale Passenger Experience
Problem: Big Data Sharing in B2B environments
Sharing Big Data Assets needs: Clearly defined and agreed common benefit Established common rules governing use, access AND benefit sharing. Organizing trust amongst group members as means to reduce risk Infrastructure supporting implementation of trust
Traditional Data Sharing model. Domain D Analyses Domain = Autonomous Organization with own administration and enforcement Domain A Domain B Domain C
Bring processing to the data. Domain D Analyses Domain A Domain B Domain C Analyses Analyses Analyses
Turntable to perform pattern searches Domain D Analyses Analyses Behavioral Pattern Simulation Domain A Domain B Domain C Analyses Analyses
Separating data from compute via high performance links. Domain D in memory analyses E.g. a 100 Gb/s link is potentially 20/80x faster when compared with a local SSD / HDD performance. Domain A Domain B Domain C
Data sharing using lightpath hub and access control Domain D $ blockchain Domain A Domain B Domain C Analyses
Digital Market Place model. Domain D In mem analyses e-market $ blockchain Domain A Domain B Domain C
Experiments at global scale using existing e- Infrastructure
Digital Market Place allowing alliance members to share data according to market rules. Digital Market Place (DMP) is a member organization as independent legal entity. Goal of the DMP is to organize trust between members wanting to gain a particular common benefit no single member can gain on its own. Members of the DMP can be a supplier or consumer of data or both. All members have equal rights within a DMP DMP is governed by a board of members in which all members participate DMP establishes a regulation consisting of market rules and the admission requirements DMP appoints a market master in charge of market operations DMP establishes a regulation for conflict settlement DMP appoints an adjudication committee Members can obtain rights (licenses) from the DMP within the framework of the DMP regulation to act in a particular defined market role. What elements of the DMP can be digitized?
Secure Digital Market Place architectural sketch National Law & Regulations Secure Digital Marketplace Member Organisation Agreement Registry Market rules Deployment Models Algorithm supplier(s) Marketplace infrastructure Deployment Specification Parameterization & authorizations Data supplier(s) Future Internet Capabilities Customer(s) Accounting & Auditing
Future Internet capabilities: Dedicated circuits offering virtually unlimited bandwidth allows big data assets to be separated from processing, offering data assets and its processing more protection. Business applications can be deployed in slices, where ownership can be expressed in a more meaningful way. Needs trust to be organized amongst organizations that like to pursue a common goal with sharing their Big Data Assets & algorithms. Digital Market Place concept and supporting infrastructure needs more research. KnuEdge LambdaFabric Importance of edge to process & store increases
Questions & acknowledgements UvA: Cees de Laat, Tom van Engers, Paola Grosso, Gleb Polevoy, Ralph Koning, Ameneh Deljoo, Ben de Graaff, Lukasz Makowski. Leiden University: Thomas Baeck TNO: Robert Meijer, Frank Fransen, NIKHEF: Tristan Suerink Ciena: Rodney Wilson, Marc Lyonais SURFnet: Erik Huizer, Gerben van Malenstein SURFsara: Anwar Osseyran, Axel Berg, Paul Melis, Maurice Bouwhuis Funding research projects: NWO: SARNET COMMIT/: SARNET Alliance NWO/STW: CIMPLO Email: leon.gommans (at) klm.com