M-PIL-3.2 Public Session
- 1. Helix Nebula – The Science Cloud with Grant Agreement 687614 is a Pre-Commercial Procurement Action funded by H2020 Framework Programme M-PIL-3.2 Public Session Next steps Andrea Chierici
- 2. Dec’18Feb’18 Ramping-up of the IaaS Resources 20k cores/2PB storage Network aggregated speed: 40Gbps Scalability Testing End User Access Combined capacity distributed across 2 Pilots Apr’18 Jun’18 14/06/2018 10k cores / 1PB storage 7k cores / 700TB storage End of Phase 3 Start of Phase 3
- 3. Quota management Procurers shared resources in order to achieve large scale tests Quota limitation may have prevented reaching important results Optimized utilization of resources (reached 80% in one case) Consolidated WLCG approach Single access point with shared resources New use-cases Each Procurer is responsible to allocate some of their capacity to all use-cases they propose In some cases of high demand the procurers voluntarily donate capacity for a limited time. 14/06/2018
- 4. Upcoming events July 9: CHEP 2018, Sofia August 28: GridKa School, Karlsruhe Hands-On session organised by KIT September 11: HNSciCloud meeting, Amsterdam Organised by SURFsara October 9-11: DI4R 2018, Lisbon October 24th: Hamburg Organised by DESY November 28th-30th: HNSciCloud meeting, CERN December 4-6: ICT 2018, Vienna Demonstrations 14/06/2018
- 5. 14/06/2018
- 6. TCO Study: PanCancer Up to 400 VMs, at least 250 VMs concurrently, 8-16 vCPUs, 16G RAM and 30 GB scratch 1 PB dataset, more than 4,000 files 5-30GB range, more than 4,000 files 100-500 MB range. Outputs ca. 92,000 files in kb range a minimal set of resources will be in constant use, and at periods of incoming data we will ramp up resource usage to a maximum Continuous deployment (will be running for the full 12 months): Compute: 7 VMs, 50 CPUs, 50GB RAM Storage: 0.5PB, not accessed frequently Network: Minimal During bursts (24-48 hours duration, 1-2x / month): Compute: ~250 VMs, 1000 vCPUs, 8,5TB RAM Storage: 0.5PB, random access with high I/O requirement Network: 10GBit/s intra-node traffic, up to 4GBit/s ingress 14/06/2018
- 7. TCO Study: Alice 3 types of jobs (workloads) within the ALICE use-case (each job uses a single core) Monte Carlo (detector simulation) low priority suitable for ‘backfilling’ unused capacity duration: 6 hours, 2 GB/core, disk 2 GB/core Network: 200MB down / 350MB up/ 0,3 Mbps Raw Data reconstruction process recently recorded data or reprocess older data Analysis Trains 14/06/2018
- 8. Intro of Vouchers short-term usage Initial voucher emission: 10 per procurer Value: 250 euro Available for any service (cpu, disk, gpu) Consumption blocked once value has been reached Possibility to export data even if credit exhausted Feedback provided by early users Expected 100 vouchers emitted in total Means of paying for IaaS services consumed by long tail of science users that will execute EGI Applications on Demand 14/06/2018