File System Needs Summary of Day #4 discussions from Dagstuhl - - PowerPoint PPT Presentation

File System Needs

Summary of Day #4 discussions from Dagstuhl Seminar #17202

Security Issues in User-level File Systems

• Stergios Anastasiadis – University of Ioannina
• Rangetable: Deamortized compactions
• Arion: Client-side journal for Ceph
• Dike: Hierarchical authentication and identification
• Security in User-level FS
• Intel’s Software Guard extensions
• Secure containers in an unsecure environment
• Limited kernel mediation for I/O
• Aerie: Storage Class Memory w/o kernel interaction

An Effort to Systematically Understand UFS Design Requirements

• Federico Padua – Universität Mainz
• Concurrency control and consistency – not much research
• Pessimistic and optimistic concurrency approaches
• POSIX standard’s take on concurrency
• Inferring false sharing with application and FS tracing

The Case for a Flexible HPC Storage Framework

• Michael Kuhn – Universität Hamburg
• Avoid specialized solutions for particular problems
• Make it easy for young researchers and students to enter FS research
• Testing ground for new technology
• JULEA – framework with pluggable interfaces for rapid prototyping
• Topside APIs: object, kv, item, posix
• Backends: posix, gio, lexos, null, mongodb
• Future work: RADOS, HDF5 VOL, etc

To FS or not to FS...

• Rob Ross – ANL
• Case study: Multi-scale modeling
• Hard problems: group membership, auth, pub/sub, and perf
• Share building blocks more effectively & reduce duplication of efforts
• Focus on new classes of data services

Building Blocks For User-level HPC Storage Systems

• Phil Carns – ANL
• Evolution of HPC storage services
• Technology transition ideas
• Building blocks: NVML, Mercury, Argobots, SSG
• Mochi project – potpourri of microservices for HPC
• New hardware and composition scenarios call for new optimization

strategies

Raghu’s summary

• Plenty of overlap in all the talks
• ^ => We all have felt/identified the same trends and challenges
• Organic growth of projects – can’t merge efforts overnight
• BUT, share/reuse building blocks, codebases, resources
• Common framework? (#4 on Kathryn’s list)