Upload
grant-chandler
View
215
Download
0
Tags:
Embed Size (px)
Citation preview
IT-DSS HEPIX Fall 2015, Alberto Pace 2
Future home directories at CERNAlberto PaceCERN, Geneva, Switzerland
With contributions from: German Cancio, Dirk Duellmann, Javier Espinal, Massimo Lamanna, Luca Mascetti, Jakub Moscicki, Andreas Peters.
IT-DSS HEPIX Fall 2015, Alberto Pace 3
Storage services at CERN• Storage of physics data
• EOS, Castor Tape Archive
• Storage of user data• DSF, AFS
• Storage of corporate documents• CDS, EDMS, Indico, Alfresco, …
120 PB, 350 million files
70 PB, 500 million files
450 TB, >3000 million files
700 TB, >800 million files
Not discussed further
IT-DSS HEPIX Fall 2015, Alberto Pace 4
Multiple roles• Castor Tape Archive
• High reliability, Low cost• Archives, physics data preservation
• High Throughput, High latency• Bulk data transfer, Integrity
Reliable and Cheap(high latency)
Fast and reliable (expensive)
Fast and cheap(Unreliable)
IT-DSS HEPIX Fall 2015, Alberto Pace 5
Multiple roles• EOS
• High scalability, Arbitrary performance• Arbitrary reliability• Designed to sustain 1 KHz access
• Main platform for Physics Analysis at CERN
Reliable and Cheap
(high latency)
Fast and reliable (expensive)
Fast and cheap
(Unreliable)
IT-DSS HEPIX Fall 2015, Alberto Pace 6
Multiple roles• AFS, DFS
• Storage of user’s and project files• Few PB, many files (billions)
• Key requirements• Simple, Scalable, Consistent, Reliable,
Available, Manageable, Flexible, Performing, Native, Cheap, and Secure.
IT-DSS HEPIX Fall 2015, Alberto Pace 7
Requirements of user’s file storage• Native online access
• As a mounted file system
• Global native access, global http access• /afs/cern.ch/user/…,
• Secure, Access Control, Encrypted• Backed up, Journaled,• Cross platform offline access• Access to physics data, unlimited storage
IT-DSS HEPIX Fall 2015, Alberto Pace 8
Several missing requirementsAFS DFS
Native online Access
Windows D CLinux C DOS X (C) DiOS D DAndroid D D
Global native online Access C (C)Offline Access, syncing (without third party tools)
Windows D CLinux D DOS X D DiOS D DAndroid D D
Web access C CAccess control C CJournaled D CEncrypted (C) (C)Access to Physics data D D
IT-DSS HEPIX Fall 2015, Alberto Pace 9
New opportunities with EOS …• Recent development for the physics data storage
infrastructure have opened new opportunities
IT-DSS HEPIX Fall 2015, Alberto Pace 10
Architecture
Client
EOS
Owncloud sync client
http OC sync protocol
(Etag, X-OC-Mtime,
If-Match, OC-FileId)
(offline access)
(online access)
Web browser access
(web access)
http, webdav,xroot
OwnCloud Web Server(implements sharing)
IT-DSS HEPIX Fall 2015, Alberto Pace 11
New opportunities with EOS …• Recent development for the physics data storage
infrastructure have opened new opportunities• Implement OwnCloud http extensions
• EOS appearing as a native Owncloud server• Empowered the use of unmodified Owncloud syncing client• Can use the Owncloud web server on top of EOS storage
• Easy sharing to unregistered identities• Read/Write sharing• Bulk download
IT-DSS HEPIX Fall 2015, Alberto Pace 12
Basic functionality• Access to the whole storage repository
• All Physics (and user) data at CERN• Could implement access to WLCG/EOS/HTTP federated storage
worldwide
• Mount EOS – Online file access• On Linux, Mac : Use FUSE or WebDAV• On Windows : Webdav (…)
• Programmatic, high performance access• Xroot (and http) libraries for all programming languages
IT-DSS HEPIX Fall 2015, Alberto Pace 13
Syncing functionality• Sync clients available for several platforms
• Can chose what to sync among the entire namespace• User / Physics / Project data DEMO
IT-DSS HEPIX Fall 2015, Alberto Pace 14
• Sync clients available for several platforms
• Can chose what to sync among the entire namespace• User / Physics / Project data
Syncing functionality
IT-DSS HEPIX Fall 2015, Alberto Pace 15
Sharing functionality• Directory-based access control lists
• Inherited from EOS service• Supports for CERN Accounts, E-groups, Roles
• Sharing functionality added by the Web Server interface• Remote users do not need an account at CERN• Read / Write access, expiry date
DEMO
IT-DSS HEPIX Fall 2015, Alberto Pace 16
Sharing functionality• Directory-based access control lists
• Inherited from EOS service• Supports for CERN Accounts, E-groups, Roles
• Sharing functionality added by the Web Server interface• Remote users do not need an account at CERN• Read / Write access, expiry date
DEMO
IT-DSS HEPIX Fall 2015, Alberto Pace 17
Usage Scenarios
Client
EOS
OwnCloud Web Server(implements sharing)Owncloud sync client
http OC sync protocol
(Etag, X-OC-Mtime,
If-Match, OC-FileId)
(offline access)
Web browser access
(online access)
(web access)
http, webdav,xroot
IT-DSS HEPIX Fall 2015, Alberto Pace 18
End-user files
Same user
EOS
OwnCloud Web Server(implements sharing)
sync clientChoose what to sync
(offline access)
(online access)
(web access)
Sharing with Third parties(web access)
Online / offlineFile Sharing(using EOSOr sync client)
Multiplatform access
IT-DSS HEPIX Fall 2015, Alberto Pace 19
Project space
EOS
OwnCloud Web Server
sync client(offline access)
(online access)
(web access)
Contributions byThird parties(web access)
(remote access using browser)
(offline user working on cached data)
(online user working remotely)
Contributions byThird parties(web access)
Multiuser access
IT-DSS HEPIX Fall 2015, Alberto Pace 20
Physics Analysis
Client
EOS
sync clientChoose what to sync(a small subset of data)
(Local Physics Data)
prototyping
batch service
http, webdav,xroot
OwnCloud Web Server(implements sharing)
Web browser access
(web access)
Analysis results
Analysis results
IT-DSS HEPIX Fall 2015, Alberto Pace 21
A new platform for analysis
Client
EOS
OwnCloud Web Serverwith Root.js
Web browser access
Virtual CPU container
http, webdav,xroot
IT-DSS HEPIX Fall 2015, Alberto Pace 22
Future of home directory services• With the availability of CERNBOX + EOS we are offering several new services,
but we have duplication of services with AFS• All AFS use cases being currently reviewed
• On-line access from desktops / laptop : sync via CERNBOX, mount via WEBDAV (evtl FUSE mount)
• World-wide remote access : Xroot, Webdav + HTTP• PLUS + BATCH Access : (permanent) FUSE mount (being investigated). HOMEDIR on
EOS• AFS project space, WEB sites hosted in AFS: Moved to CERNBOX / EOS (being
investigated)• Other cases (SW build & distribution, …) : local builds, CVMFS
• Currently asking the user community to report • use cases for which end-users have concerns• Other use cases not on the list that may require alternate/additional solutions or
investigations
IT-DSS HEPIX Fall 2015, Alberto Pace 23
Questions ?