Globus: Enabling the Open Storage Network
1 like444 views
The Open Storage Network (OSN) aims to provide a sustainable, low-cost distributed storage cloud for the NSF research community, leveraging existing high-speed networks and promoting data sharing. The OSN features a standardized petascale storage infrastructure designed to facilitate interdisciplinary research, with scalable units referred to as 'OSN pods' that have specific technical requirements and specifications. Funded by NSF and the Schmidt Foundation, OSN focuses on simplifying deployment and management while enhancing accessibility to publicly-funded research datasets.
Globus: Enabling the Open Storage Network
- 1. Globus: Enabling the Open Storage Network GlobusWorld 2019 Brian Mohr Made possible by grants from the NSF and the Schmidt Foundation NSF grants 1747552 1747493 1747507 1747490 1747483 1836357
- 2. The Open Storage Network: Mission Statement The mission of OSN is to provide a low-cost, high-quality, sustainable, distributed storage cloud for the NSF research community.
- 3. Research Cyberinfrastructure Today Shared Resource (XSEDE, PRAC) Standardized NSF-Funded 200+ universities with 40/100Gb Connectivity Standardized NSF-Funded Largely Balkanized No Standards Requirement No CI Funding Computation Networking Storage
- 4. The Open Storage Network: Cyberinfrastructure Goals q Leverage existing NSF-funded high-speed network connectivity q Establish a standard national petascale storage infrastructure q Promote sharing of publicly-funded research datasets q Facilitate interdisciplinary research (searchable metadata) The third pillar…
- 5. OSN Federation Design Objectives q Scalable: uniform hardware architecture across all OSN sites q Efficient: centralized remote provisioning and monitoring q Manageable: distributed lights-out data center q Resilient: offline site ≠ loss of access to data (cross-site replication) q Sustainable: minimal site-local staff overhead Keep it simple…
- 6. OSN Pod Design Objective: “Scalable Unit” q Capacity: 1 petabyte usable object storage q Performance: 40Gb sequential throughput q Ease of Procurement: an OSN-optimized vendor SKU q Economical: $140,000 hardware budget q Ease of Adoption: plug-n-play appliance model Keep it simple…
- 7. OSN Scalable Unit – Technical Spec 8 Server Nodes Five 4U Data Nodes | Three 1U Monitor/DTN Nodes 1.44 PB Raw Storage 8T HDDs | 7200 RPM 12Gb SAS | 36 Disks per Data Node High-Speed Network: 100GbE ToR Switch 40 or 100Gb I2 Uplink | 50GbE Cluster Interconnect Remote Management: 1GbE, KVM ToR Switches OOB IPv4 | IPMI | Console | Switched Outlets
- 8. OSN Pod Physical Site Requirements q Floor space for one 30-inch wide rack q A/B power (current configuration: 6kW max) q Dual fiber uplink to 40/100Gb network infrastructure q Out-of-band network access for remote “pod” provisioning q Allocate IP address blocks (high-speed: /27; OOB: 3x IPv4) q That’s it! Minimize barriers to adoption…
- 9. OSN Software Stack: Globus q Authentication infrastructure q GridFTP file transfer software q S3 interface to Ceph object storage Leverage existing Globus features…
- 10. OSN Software Stack: Globus Extensions q Dataset Ownership q Dataset Access Authorization q Dataset Locality q Dataset Replication q Dataset Aging q Dataset Tags (Searchable Catalog!) Develop OSN-specific metadata/policy engine…
- 11. OSN Pod/Appliance Operations Monitoring Globus Engine Provisioning
- 12. OSN Prototype Deployment Sites Northeastern Storage Exchange San Diego Supercomputing Center University of Illinois Renaissance Computing Institute Northwestern University Johns Hopkins University Funded by NSF Funded by Schmidt Foundation
- 13. OSN Scaled-out Deployment (Projected) Assumption: one or more OSN Pods per 40/100Gb NSF Campus Cyberinfrastructure Site.