SILECS/SLICES - Super Infrastructure for Large-Scale Experimental Computer Science
•
0 likes•239 views
The aim of the SILECS and SLICES projects is to design and build a large infrastructure for experimental research on various aspects of distributed computing, from small connected objects to the large data centres of tomorrow. This infrastructure will allow end-to-end experimentation with software and applications at all levels of the software layers, from event capture (sensors, actuators) to data processing and storage, to radio transmission management and dynamic deployment of edge computing services, enabling reproducible research on all-point programmable networks, ... SILECS is the french node of a european infrastructure called SLICES.
Report
Share
SILECS/SLICES - Super Infrastructure for Large-Scale Experimental Computer Science
- 1. SILECS/SLICES Super Infrastructure for Large-Scale Experimental Computer Science (Almost) everything you wanted to know about SILECS/SLICES but didn't dare to ask F. Desprez – Inria/LIG, S. Fdida – Sorbonne University F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr INRIA, CNRS, RENATER, IMT, Sorbonne Université, Université Grenoble Alpes, Université Lille 1, Université Lorraine, Université Rennes 1, Université Strasbourg, Université fédérale de Toulouse, ENS Lyon, INSA Lyon, … http://www.silecs.net/
- 2. The Discipline of Computing: An Experimental Science The reality of computer science - Information - Computers, networks, algorithms, programs, etc. Studied objects (hardware, programs, data, protocols, algorithms, networks) are more and more complex Modern infrastructures • Processors have very nice features: caches, hyperthreading, multi-core, … • Operating system impacts the performance (process scheduling, socket implementation, etc.) • The runtime environment plays a role (MPICH ≠ OPENMPI) • Middleware have an impact • Various parallel architectures that can be heterogeneous, hierarchical, distributed, dynamic F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 3. Good Experiments A good experiment should fulfill the following properties – Reproducibility: must give the same result with the same input – Extensibility: must target possible comparisons with other works and extensions (more/other processors, larger data sets, different architectures) – Applicability: must define realistic parameters and must allow for an easy calibration – “Revisability”: when an implementation does not perform as expected, must help to identify the reasons F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 4. SILECS/SLICES Motivation • Exponential improvement of – Electronics (energy consumption, size, cost) – Capacity of networks (WAN, wireless, new technologies) • Exponential growth of applications near users – Smartphones, tablets, connected devices, sensors, … – Large variety of applications and large community • Large number of Cloud facilities to cope with generated data – Many platforms and infrastructures available around the world – Several offers for IaaS, PaaS, and SaaS platforms – Public, private, community, and hybrid clouds – Going toward distributed Clouds (FOG, Edge, extreme Edge) F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 5. SLICES – ESFRI Call (Sept. 2020) • Core partners • Belgium • Cyprus • France • Greece • Hungary • Italy • Luxembourg • Netherland • Norway • Poland • Spain • Switzerland F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr • Under discussion • Sweden • GIANT and national NRENs
- 6. SILECS – PIA-3/EQUIPEX+ call (June 2020) • Core partners • Inria • CNRS • IMT • Université fédérale de Toulouse • Université Strasbourg • Université Grenoble Alpes • Université de Lille • Université de Lorraine • Sorbonne Université • Renater • Eurecom • ENS Lyon • INSA de Lyon F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 7. Envisioned Architecture F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 8. SILECS/GRID’5000 • Testbed for research on distributed systems • Born in 2003 from the observation that we need a better and larger testbed • HPC, Grids, P2P, and now Cloud computing, and BigData systems • A complete access to the nodes’ hardware in an exclusive mode (from one node to the whole infrastructure) • Dedicated network (RENATER) • Reconfigurable: nodes with Kadeploy and network with KaVLAN • Current status • 8 sites, 36 clusters, 838 nodes, 15116 cores • Diverse technologies/resources (Intel, AMD, Myrinet, Infiniband, two GPU clusters, energy probes) • Some Experiments examples • In Situ analytics • Big Data Management • HPC Programming approaches • Network modeling and simulation • Energy consumption evaluation • Batch scheduler optimization • Large virtual machines deployments F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr https://www.grid5000.fr/
- 9. SILECS/ FIT F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr FIT-IoT-LAB • 2700 wireless sensor nodes spread across six different sites in France • Nodes are either fixed or mobile and can be allocated in various topologies throughout all sites Sophia Lyon FIT-CorteXlab: Cognitive Radio Testbed 40 Software Defined Radio Nodes (SOCRATE) FIT-R2Lab: WiFi mesh testbed (DIANA) https://fit-equipex.fr/ https://www.iot-lab.info/hardware/ Providing Internet players access to a variety of fixed and mobile technologies and services, thus accelerating the design of advanced technologies for the Future Internet
- 10. Data Center Portfolio Targets ● Performance, resilience, energy-efficiency, security in the context of data-center design, Big Data processing, Exascale computing, AI, etc. Hardware ● Servers: x86, ARM64, POWER, accelerators (GPU, FPGA), … ● AI dedicated servers ● Edge computing micro datacenters ● Networking: Ethernet (10G, 40G), HPC networks (InfiniBand, Omni-Path), … ● Storage: HDD, SSD, NVMe, both in storage arrays and clusters of servers, … Experimental support ● Bare-metal reconfiguration ● Large clusters ● Integrated monitoring (performance, energy, temperature, network traffic) F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 11. Wireless Portfolio Targets • Performance, security, safety and privacy-preservation in complex sensing environment, �� Performance understanding and enhancement in wireless networking, • Target applications: smart cities/manufacturing, building automation, standard and interoperability, security, energy harvesting, health care Hardware • Software Defined Radio (SDR), NB-IoT, 5G, BLE, Thread • Wireless Sensor Network (IEEE 802.15.4), • LoRa/LoRaWAN, … Experimental support • Bare-metal reconfiguration • Large-scale deployment (both in terms of densities and network diameter) • Different topologies with indoor/outdoor locations • Mobility-enabled with customized trajectories • Anechoic chamber • Integrated monitoring (power consumption, radio signal, network traffic) F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 12. Outdoor IOT testbed • IoT is not limited to smart objects or indoor wireless sensors (smart building, industry 4.0, ….) • Smart cities need outdoor IoT solutions • Outdoor smart metering • Outdoor metering at the scale of a neighborhood (air, noise smart sensing, ….) • Citizens and local authorities are more and more interested by outdoor metering • Controlled outdoor testbed • (Reproducible) polymorphic IoT: support of multiple IoT technologies (long, middle and short range IoT wireless solutions) at the same time on a large scale testbed • Agreement and support of local authorities • Deployment in Strasbourg city (500000 citizens, 384 km2) F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 13. An experiment outline • Discovering resources from their description • Reconfiguring the testbed to meet experimental needs • Monitoring experiments, extracting and analyzing data • Controlling experiments: API F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 14. Plans for SILECS/SLICES: Testbed Services ● Provide a unified framework that (really) meets all needs ○ Make it easier for experimenters to move for one testbed to another ○ Make it easy to create simultaneous reservations on several testbeds (for cross- testbeds experiments) ○ Make it easy to extend SILECS/SLICES with additional kinds of resources ● Factor testbed services ○ Services that can exist at a higher level, e.g. open data service, for storage and preservation of experiments data ○ In collaboration with Open Data repositories such as OpenAIRE/Zenodo ○ Services that are required to operate such infrastructures, but add no scientific value ○ Users management, usage tracking F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 15. Services & Software Stack F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr Built from already functional solutions
- 16. The GRAIL F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 17. Some recent experiments examples • QoS differentiation in data collection for smart Grids, J. Nassar, M. Berthomé, J. Dubrulle, N. Gouvy, N. Mitton, B. Quoitin • Damaris: Scalable I/O and In-situ Big Data Processing, G. Antoniu, H. Salimi, M. Dorier • Frequency Selection Approach for Energy Aware Cloud Database, C. Guo, J.-M. Pierson • Distributed Storage for a Fog/Edge infrastructure based on a P2P and a Scale-Out NAS, B. Confais, B. Parrein, A. Lebre • FogIoT Orchestrator: an Orchestration System for IoT Applications in Fog Environment, B. Donassolo, I. Fajjari, A. Legrand, P. Mertikopoulos F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 18. QoS differentiation in data collection for smart Grids • Data collection with different QoS requirements for Smart Grid applications • Traditional approach • Use of standard RPL protocol which offers overall good performance but no QoS differentiation based on application • Solution • Use a dynamic objective function • FIT IoT LAB as a validation testbed • Access to 67 sensor nodes with IoT features remotely • Customizable environment and tools (data size and rate, consumption measure, clock, etc) • Repeat the experiments and compare to alternate approaches with the same environment • The results show that based on the service requested, data from different applications follow different paths, each meeting expected requirements • FIT IoT LAB helped validate the approach to go further with standardization F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr Multiple Instances QoS Routing In RPL: Application To Smart Grids – J. Nassar, M. Berthomé, J. Dubrulle, N. Gouvy, N. Mitton, B. Quoitin – MDPI Sensors, July 2018
- 19. Damaris • Scalable, asynchronous data storage for large-scale simulations using the HDF5 format (HDF5 blog at https://goo.gl/7A4cZh) • Traditional approach • All simulation processes (10K+) write on disk at the same time synchronously • Problems: 1) I/O jitter, 2) long I/O phase, 3) Blocked simulation during data writing • Solution • Aggregate data in dedicated cores using shared memory and write asynchronously • Grid’5000 used as a testbed – Access to many (1024) homogeneous cores – Customizable environment and tools – Repeat the experiments later with the same environment saved as an image • The results show that Damaris can provide a jitter-free and wait-free data storage mechanism • G5K helped prepare Damaris for deployment on top supercomputers (Titan, Pangea (Total), Jaguar, Kraken, etc.) F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr … https://project.inria.fr/damaris/
- 20. Frequency Selection Approach for Energy Aware Cloud Database • Objective: Study the energy efficiency of cloud database systems and propose a frequency selection approach and corresponding algorithms to cope with resource proposing problem F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr Frequency Selection Approach for Energy Aware Cloud Database, C. Guo, J.-M. Pierson. In Proc. SBAC-PAD, 2018. Relationship between Request Amount and Throughput • Contribution: Propose frequency selection model and algorithms. • Propose a Genetic Based Algorithm and a Monte Carlo Tree Based Algorithm to produce the frequencies according to workload predictions • Propose a model simplification method to improve the performance of the algorithms • Grid5000 usage • A cloud database system, Cassandra, was deployed within a Grid’5000 cluster using 10 nodes of Nancy side to study the relationship between system throughput and energy efficiency of the system • By another benchmark experiment, the migration cost parameters of the model were obtained
- 21. Distributed Storage for a Fog/Edge infrastructure based on a P2P and a Scale-Out NAS • Objective • Design of a storage infrastructure taking locality into account • Properties a distributed storage system should have: data locality, network containment, mobility support, disconnected mode, scalability • Contributions • Improving locality when accessing an object stored locally coupling IPFS and a Scale- Out NAS • Improving locality when accessing an object stored on a remote site using a tree inspired by the DNS • Experiments • Deployment of a Fog Site on the Grid’5000 testbed and the clients on the IoTLab platform • Coupling a Scale-Out NAS to IPFS limits the inter-sites network traffic and improves locality of local accesses • Replacing the DHT by a tree mapped on the physical topology improves locality to find the location of objects • Experiments using IoTlab and Grid’5000 are (currently) not easy to perform F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr An Object Store Service for a Fog/Edge Computing Infrastructure based on IPFS and Scale-out NAS, B. Confais, A. Lebre, and B. Parrein (May 2017). In: 1st IEEE International Conference on Fog and Edge Computing - ICFEC’2017.
- 22. FogIoT Orchestrator: an Orchestration System for IoT Applications in Fog Environment F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr • Objective • Design a Optimized Fog Service Provisioning strategy (O-FSP) and validate it on a real infrastructure • Contributions • Design and implementation of FITOR, an orchestration framework for the automation of the deployment, the scalability management, and migration of micro-service based IoT applications • Design of a provisioning solution for IoT applications that optimizes the placement and the composition of IoT components, while dealing with the heterogeneity of the underlying Fog infrastructure • Experiments • Fog layer is composed of 20 servers from Grid’5000 which are part of the genepi cluster, Mist layer is composed of 50 A8 nodes • Use of a software stack made of open-source components (Calvin, Prometheus, Cadvisor, Blackbox exporter, Netdata) • Experiments show that the O-FSP strategy makes the provisioning more effective and outperforms classical strategies in terms of: i) acceptance rate, ii) provisioning cost, and iii) resource usage FogIoT Orchestrator: an Orchestration System for IoT Applications in Fog Environment, B. Donassolo, I. Fajjari, A. Legrand, P. Mertikopoulos.. 1st Grid’5000-FIT school, Apr 2018, Sophia Antipolis, France. 2018.
- 23. SILECS: Based upon Two Existing Infrastructures • FIT – Providing Internet players access to a variety of fixed and mobile technologies and services, thus accelerating the design of advanced technologies for the Future Internet – 4 key technologies and a single control point: IoT-Lab (connected objects & sensors, mobility), CorteXlab (Cognitive Radio), R2Lab (anechoic chamber), Cloud technology including OpenStack, Network Operations Center – 9 sites (Paris (2), Evry, Rocquencourt, Lille, Strasbourg, Lyon, Grenoble, Sophia Antipolis) • Grid’5000 – A scientific instrument for experimental research on large future infrastructures: Clouds, datacenters, HPC Exascale, Big Data infrastructures, networks, etc. – 8 sites, > 15000 cores, with a large variety of network connectivity and storage access, dedicated interconnection network granted and managed by RENATER • Software stacks dedicated to experimentation • Resource reservation, disk image deployment, monitoring tools, data collection and storage F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 24. Proxy location selection in industrial IoT • Distributed data collection with low latency in Industrial context • Traditional approach • Improving data routing by selecting quicker links • Deploying enhanced edge-nodes for fog computing • Solution • Dynamically select sensor nodes to act as proxys and get the information closer to consuming nodes. • FIT IoT LAB as a validation testbed • Access to 95 sensor nodes with IoT features remotely • Customizable environment and tools (sniffer, consumption measure, etc) • Repeat the experiments later and compare to alternate approaches with the same environment • The results show that latency is much reduced • FIT IoT LAB helped validate the approach before real costly deployment F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr Performance Analysis of Latency-Aware Data Management in Industrial IoT Networks, T.P. Raptis, A. Passarella, M. Conti - MDPI Sensors, 2018, 18(8), 2611
- 25. KerA: Scalable Data Ingestion for Stream Processing • Goal: increase ingestion and processing throughput of Big Data streams • Dynamic partitioning and lightweight stream offset indexing • Higher parallelism for producers and consumers • Grid’5000 Paravance cluster used for development and testing • Customized OS image and easy deployment • 128GB RAM and 16 CPU cores • 10Gb networking • Next steps: KerA* unified architecture for stream ingestion and storage • Support for records, streams and objects • Collaborations • INRIA, HUAWEI, UPM, BigStorage F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr KerA: Scalable Data Ingestion for Stream Processing, O.-C. Marcu, A. Costan, G. Antoniu, M. Pérez-Hernández, B. Nicolae, R. Tudoran, S. Bortoli. In Proc. ICDCS, 2018. KerA vs Kafka: up to 4x-5x better throughput
- 26. Conclusions • SLICES: new infrastructure for experimental computer science and future services in Europe • SILECS: new infrastructure in France based on two existing instruments (FIT and Grid’5000) • Big challenges ! • Design a software stack that will allow experiments mixing both kinds of resources at the European level while keeping reproducibility level high • Keep the existing infrastructures up while designing and deploying the new one • Keep the aim of previous platforms (their core scientific issues addressed) – Scalability issues, energy management, … – IoT, wireless networks, future Internet – HPC, big data, clouds, virtualization, deep learning, ... • Address new challenges – IoT and Clouds – New generation Cloud platforms and software stacks (Edge, FOG) – Data streaming applications – Locality aware resource management – Big data management and analysis from sensors to the (distributed) cloud – Mobility – Next generation wireless – … • Next steps – PIA-3 (Equipements structurants pour la recherche/EQUIPEX+) and ESFRI F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr
- 27. Thanks, any questions ? F. Desprez - SILECS/SLICES - Frederic.Desprez@inria.fr http://www.silecs.net/ https://www.grid5000.fr/ https://fit-equipex.fr/