FEDERICA Federated e-Infrastructure supporting research experiments on new Internet architectures and protocols Status and Experience Mauro Campanella - GARR TNC Vilnius 1 st June 2010

2 TNC2010, VILNIUS, 2nd June Agenda –FEDERICA brief overview, framework & architecture –Infrastructure status –Service model, reproducibility –Initial analysis of users –Challenges and conclusions

3 TNC2010, VILNIUS, 2nd June FEDERICA at a glance What:European Community co-funded project in its 7 th Framework Program in the area “Capacities - Research Infrastructures”, 3.7 MEuro EC contribution, 461 Person Months When:1 st January October 2010 (30 months) Who:20 partners, based on stakeholders on network research and management: 11 National Research and Education Networks, DANTE (GÉANT), TERENA, 4 Universities, Juniper Networks, 1 small enterprise (MARTEL), 1 research centre (i2CAT) - Coordinator: GARR (Italian NREN) Where:Europe-wide e-Infrastructure, open to external connections

4 TNC2010, VILNIUS, 2nd June FEDERICA Deploy and manage an e-Infrastructure (FEDERICA) based on virtualization in both computers and network elements as a fundamental tool/playground for researchers on current and Future Internet, its uses and technologies. It allows researchers a complete control of their set of resources (a “slice”), poses a minimum number of cons- traints and enables disruptive experiments at all communication layers over a realistic substrate. Particular focus is placed in reproducibility of the experiments and in the avoidance of complexity. Router/Switch Host for Virtual nodes

5 TNC2010, VILNIUS, 2nd June FEDERICA Framework The infrastructure requirements brought to 2 key framework choices for the infrastructure, at the core of design: 1.The simultaneous presence of computing and network physical resources. These resources form the substrate of the infrastructure. 2.The use of virtualization technologies applied both to computing and network resources. Virtualization will allow creating virtual, un-configured resources. Framework is compatible with the Infrastructure as a Service (IaaS) used in “clouds”, including network layers and it may offer other services (e.g. Platform aaS and Software aaS).

6 TNC2010, VILNIUS, 2nd June FEDERICA - Current Architecture Two levels: 1.The virtualization substrate. The physical infrastructure which contains all the hardware and software capable to create the virtual resources. 2.The level containing all virtual infrastructures (slices). Each slice contains a set of virtual resources and their network topology. Two elementary resource entities: 1.Connectivity. In form of a bit pipe (point to point circuit) with or without capacity guarantees and with or without a data link protocol. 2.A computing element and its attributes, offering the equivalent of a computer hardware containing at least RAM, CPU and one network interface, mass storage is optional. The computing element is capable of hosting various operating systems and perform also functionalities (e.g. routing). This differs from some “Clouds” where storage is considered an independent entity.

7 TNC2010, VILNIUS, 2nd June Router/Switch Host for Virtual nodes Raw Ethernet 1Gbps (Fiber later) The Resources Architectural layers (pictorial) The first order slices are vertical to indicate that there is no hierarchy between them. Virtual resources in different slices may be created in the same physical resource.

8 TNC2010, VILNIUS, 2nd June Agenda –FEDERICA brief overview, framework & architecture –Infrastructure status –Service model, reproducibility –Initial analysis of users –Challenges and conclusions

9 TNC2010, VILNIUS, 2nd June Infrastructure Status Up and operational (with users) All links are 1 Gbps Ethernet GARR IT DFN DE CESNET CZ SWITCH CH Red.es ES GRNET ICCS GR Hungarnet HU PSNC PL HEAnet IE i2CAT ES KTH SE NORDUN ET SUNET FCCN PT Each PoP hosts one or more large PC and a network switch/router

10 TNC2010, VILNIUS, 2nd June BGP Peering 1 Gbps Ethernet RS-232 Legenda FEDERICA V-Node Out-Of-Band Terminal server Other FEDERICA PoPs NREN Production Network Notes: -Each V-Node has 7+1network interfaces at 1 Gbps Ethernet -OOB is not mandatory FEDERICA V-Node FEDERICA Juniper MX480 FEDERICA Core PoP GEANT INTERNET

11 TNC2010, VILNIUS, 2nd June The Core Substrate - HW and IP Single IP Autonomous System: AS :(public, no transit, the 4 core nodes peer with local NRENs which announce the AS to GN2 and Global Internet) active IP v4 active IPv6 (under tests) FEDERICA substrate NRENs and Global Internet Router / Switch: Juniper MX480, Dual CPU, 1 line card with 32 ports at 1Gb Ethernet. Virtual and logical routing, MPLS, VLANs, IPv4, IPv6, 2 of the 4 line cards have hardware QoS packet- based capabilities V-Nodes: each is a 2 x Quad core 2GHz, 32GB RAM, /100/10 Ethernet NICs, 2x500GB disks, VMware ESXi 3.5

12 TNC2010, VILNIUS, 2nd June Monitoring: the whole system 12 G3 system from CESNET, including Nagios, enhanced (e.g. for QinQ). Monitors all circuits and nodes (> 50K entities monitored), all slices Dynamic interval time, usually in ‘relaxed’ mode (longer delay between measures)

13 TNC2010, VILNIUS, 2nd June CPUs each 1775 MHz = 14,2 GHz

14 TNC2010, VILNIUS, 2nd June Substrate behavior Network circuits and nodes can use all available capacity. Systematic measurement are ongoing and results will be published for the substrate.

15 TNC2010, VILNIUS, 2nd June Substrate behavior - V-Nodes The current set-up of the slices ensures there is no overbooking and behaviour of V-Nodes is stable.

16 TNC2010, VILNIUS, 2nd June Agenda –FEDERICA brief overview, framework & architecture –Infrastructure status –Service model, reproducibility –Initial analysis of users –Challenges and conclusions

17 TNC2010, VILNIUS, 2nd June The basic service model The FEDERICA basic service is the creation of “virtual infrastructures” with full user control. It is only lightly dependent on the underlying architecture. The project made a choice to have two characteristics of the service that imply manual intervention in the initial phase of user access : - the User Policy Board to accept, register, prioritize and counsel the users’ proposals (AAI is included here) - the reproducibility requirement which (up to now) is a manual mapping from physical to virtual resources. Advanced brokers / technologies / over provisioning may overcome this limit (see ongoing work in clouds). The user receives then full control of the slice.

18 TNC2010, VILNIUS, 2nd June Virtual Internet Virtual to Physical resource mapping The step to map to the physical resources the virtual resources has been chosen to be performed manually in FEDERICA. The motivations are: - ensure reproducibility when requested - optimization of the use of the infrastructure.

19 TNC2010, VILNIUS, 2nd June FEDERICA versus other projects Project: Feature: Onelab2/ Planetlab EmulabFEDERICA Allowed Operating system fixed User’s choice (almost any) Control of lower layers Noemulation control down to raw Ethernet IPMandatory used to connect then emulation Used to connect, then not needed Choice of physical delay and capacity No, unless in SWemulation Yes (up to 1 Gb) reproducibility guarantees NoemulationYes User access limitations Almost none, at any time Almost none, at any time initially regulated by a User Policy Board Costlimitednone Scalabilitymedium/largelargemedium/limited

20 TNC2010, VILNIUS, 2nd June Offering “Slices” for “any” Research Using Virtualization technologies the FEDERICA e-Infrastructure creates “slices” composed by virtual resources (circuits, nodes, routers) FEDERICA substrate (only core is shown) Global Internet The slices are configured according to users’ requests Possible use cases: - new routing protocols - behavior on the network of distributed applications - Inter-domain services

21 TNC2010, VILNIUS, 2nd June Reproducibility FEDERICA ensures reproducibility of the behaviour of a single virtual resource as a key requirement for quantitative evaluations of new ideas through experimentation. As an example, a virtual circuit may not be capable of offering a constant, fixed amount of bit per second, and a virtual computer image may not provide a constant CPU percentage and/or a constant disk access capacity. The project does not instead aims at providing QoS for sets of connected resources, or entire slices, as this is a research area still. Most of the projects/experiments do not request reproducibility in those cases a ‘good’ resources behaviour is provided anyway (see next slide).

22 TNC2010, VILNIUS, 2nd June Reproducibility - How to The project engineered the substrate so that: -there is a high level of network physical meshing at 1Gbps -router/switches perform at line rate -CPUs assist virtualization in HW -Each V-Node has -a large number of physical network interfaces -at 32 GB RAM (4 GB for each core) -multiple disks in separate channels Rules are followed when creating resources(strictly when reproducibility is requested, preferably in the other cases): -only one virtual host runs in each core -RAM is statically partitioned -Each virtual host receives a dedicated physical network interface -The circuits are never over-subscribed

23 TNC2010, VILNIUS, 2nd June Agenda –FEDERICA brief overview, framework & architecture –Infrastructure status –Service model, reproducibility –Initial analysis of users –Challenges and conclusions

24 TNC2010, VILNIUS, 2nd June Resource Usage per slice R < 10: Feasibility study, 40 < R < 100: Validation, 10 < R < 40: Functionality,100 < R: Scalability Resources (R) Requested

25 TNC2010, VILNIUS, 2nd June Slice requested time duration Slice Lease Time

26 TNC2010, VILNIUS, 2nd June Motivation of research –ICT research on networking –ICT research on applications/services –ICT research on both applications/services and networking –Applied-ICT research on applications/services (eg medecine)

27 TNC2010, VILNIUS, 2nd June Rationale od the use of FEDERICA –Economical: Renting vs. Buying –Opportunity: Exploit large number of resources and/or fast provisioning –Technical: Exploit new functionalities –Federated: Exploit interconnection of infrastructures

28 TNC2010, VILNIUS, 2nd June Agenda –FEDERICA brief overview, framework & architecture –Infrastructure status –Service model, reproducibility –Initial analysis of users –Challenges and conclusions

29 TNC2010, VILNIUS, 2nd June Research areas (Challenges) There are a some key challenges for the evolution of the project and tightly connected to virtualization : 1.“Real” versus “Virtual”, i.e. reproducibility/QoS 2.Virtualization service definition and automation of procedures, in particular resource mapping. 3.Federation 4.Complexity (virtualization may be recursive)

30 TNC2010, VILNIUS, 2nd June User Support Challenges There are key challenges to provide a smooth and easily accessible service to researchers. Some lesson learned from experience: 1.The user access should be as simple and convenient as much possible (reduce formal needs, automate proposal, provide examples and instruction). Needs substantial programming effort. 2.Basic tasks should be automated both for the noc and for the users (circuit and V-Node provisioning, configuration, sample - preconfigured nodes, …) 3.Support of research is different from support of a production environment and more time-consuming (more complex questions and requests)

31 TNC2010, VILNIUS, 2nd June Conclusions For Future (and present) Internet research and evolution, FEDERICA- like infrastructures, enable innovative paths, exploiting virtualization both in computers and networks. FEDERICA creates virtual infrastructures, behaving almost like real ones, which can be seen as generalised, “cloud” infrastructures. The user access is simple from any place connected to Internet. Universities, Private Enterprises even schools may benefit from such ready-to-use environment. Many challenges and research areas are still to be explored, analyzed to create a common set of standards. Virtualization facilitates and draws a smoother path to federation too. FEDERICA can start federating with other facilities.

32 TNC2010, VILNIUS, 2nd June More Information Main source of information is (all deliverables and documents, user access portal redirection) 1.Dae Young Kim, Laurent Mathy, Mauro Campanella, Rick Summerhill, James Williams, Shinji Shimojo, Yasuichi Kitamura, Hideaki Otsuki, "Future Internet: Challenges in Virtualization and Federation," aict, pp.1-8, 2009 Fifth Advanced International Conference on Telecommunications, 2009, Venice/Mestre, Italy, May 24-May 28, ISBN: P. Sezgedi, S. Figuerola, M. Campanella, V. Maglaris, C. Cervello-Pastor: "With Evolution for Revolution: Managing FEDERICA for Future Internet Research", IEEE Communications Magazine Vol.47 No.7 pp , July M.Campanella, "The FEDERICA Project: creating cloud infrastructures", In Proceedings of Cloudcomp 2009 (CloudComp), October 19-21, 2009, Munich, Germany. ISBN , 4.M.Campanella, V.Maglaris, M.Potts “Virtual Infrastructures in Future Internet”, in “Towards the Future Internet - Emerging Trends from European Research”, IOS press, April 2010, ISBN

33 TNC2010, VILNIUS, 2nd June FEDERICA Consortium National Research & Education Networks CESNETCzech Rep. DFNGermany FCCNPortugal GARR (coordinator)Italy GRNETGreece HEAnetIreland NIIF/HUNGARNETHungary NORDUnetNordic countries PSNCPoland Red.esSpain SWITCHSwitzerland Small Enterprise Martel ConsultingSwitzerland NRENs organizations TERENA The Netherlands DANTE United Kingdom Universities - Research Centers i2CATSpain KTHSweden NTUA (ICCS)Greece UPCSpain PoliTOItaly System vendors Juniper NetworksIreland

34 TNC2010, VILNIUS, 2nd June Thank you for your attention