Transcript pptx - XRootD

XRootD
LSST Data Management
Database Architecture Review
SLAC National Accelerator Laboratory
August 6-7, 2013
Andrew Hanushevsky, SLAC
http://xrootd.org
A Bit Of XRootD History
1997 – Objectivity, Inc. collaboration

Design & Development to scale Objectivity/DB


First attempt to use commercial DB for Physics data
Successful but problematical
2001 – BaBar decides to use root framework vs Objectivity

Collaboration with INFN, Padova & SLAC created


Design & develop high performance data access system
Work based on what we learned with Objectivity
2003 – First deployment of XRootD system at SLAC
2013 – Wide deployment in LHC & Astrophysics


ALICE, ATLAS, CMS, EXO, Fermi/GLAST, LSST; among others
Protocol also available in dCache, DPM, and EOS
August 6-7, 2013
LSST Database Architecture Review
‹#›
What Is XRootD?
A system for scalable cluster data access
xrootd
cmsd
Data Access
Data Clustering
Not a file system & not just for file systems
If you can write a plug-in you can cluster it

The essential key for Qserv
3
August 6-7, 2013
LSST Database Architecture Review
‹#›
XRootD Plug-In Architecture
Authentication
Protocol Driver
krb5 sss x.509 …
Any n protocols
Protocol
cms http xroot …
Authorization
Entity Names
Logical File System
Logical
System
(OFSFile
Plugin)
StorageQserv
System
Scalable
Interface
dpmService
sfs sql
…
HDFS gpfs
Lustre
UFS, …
Plus
mySQL
Clustering
Clustering
(cmsd)
(cmsd)
August 6-7, 2013
LSST Database Architecture Review
‹#›
4
The Original Challenge
The physics analysis regime “problem”
Write once read many times access mode
 Small block sparse random I/O
 Hundreds of servers of uneven reliability
 Thousands parallel batch jobs
 100’s of thousands file sessions

The LSST Challenge
August 6-7, 2013
LSST Database Architecture Review
‹#›
XRootD Synergistic Solution
Minimize latency
Parallelizable protocol, file sessions, lockless I/O, sticky threads
Minimize hardware requirements
Short code paths, compact objects
Cache aware members
Minimal data movement
No cross-thread data sharing
Minimize human cost
Single configuration file
Cookie cutter installation
No database needed
Real time node change,
Native FS admin tools
Maximize scaling
Result
<7 us overhead & <100 MB footprint & unlimited servers
August 6-7, 2013
LSST Database Architecture Review
‹#›
xrootd
XRootD B64 Scaling
cmsd
xrootd
Replicated & Load Balanced
xrootd
xrootd
cmsd
Manager
(Root Node)
cmsd
cmsd
xrootd
641 =
64
cmsd
xrootd
cmsd
642
=
4096
xrootd
xrootd
xrootd
xrootd
cmsd
cmsd
cmsd
cmsd
xrootd
xrootd
cmsd
cmsd
643 =
262144
Supervisors
(Interior Nodes)
xrootd
xrootd
cmsd
cmsd
644 = 16777216
xrootd
xrootd
xrootd
xrootd
cmsd
cmsd
cmsd
cmsd
Data Servers
(Leaf Nodes)
Qserv worker nodes reside only at the leaves
Each identified by a dynamic set of arbitrary names
August 6-7, 2013
LSST Database Architecture Review
‹#›
xrootd
WYSIWYG Scalable Access
open()
redirect
open()
redirect
open()
Request routed
to an alternate
node exporting
same logical name
cmsd
Client
xrootd
cmsd
xrootd
641 =
64
cmsd
xrootd
cmsd
642
=
Exponentially Parallel Query
For Logical Endpoint Name
Routing Paths Cached
At Each Router Node
4096
xrootd
xrootd
xrootd
xrootd
cmsd
cmsd
cmsd
cmsd
Request routing is very different from traditional data management models
This implements a structured network of request routers
Capable of automatically recovering from adverse conditions
Much like internet routing
August 6-7, 2013
LSST Database Architecture Review
‹#›
Conclusion
A facile, flexible, and sound system
 Applicable to a wide variety of problems
LGPL open-source
Managed by the XRootD collaboration

SLAC, CERN, Duke, JINR, UCSD, & UNL (fall)
More at http://xrootd.org/
August 6-7, 2013
LSST Database Architecture Review
‹#›
Acknowledgements
Current Contributors









ATLAS: Doug Benjamin, Patrick McGuigan, Ilija Vukotic
CERN: Lukasz Janyst, Andreas Peters, Justin Salmon
Fermi: Tony Johnson
JINR: Danila Oleynik, Artem Petrosyan
Root: Gerri Ganis, Bertrand Bellenet, Fons Rademakers
SLAC: Andrew Hanushevsky, Wilko Kroeger, Daniel Wang, Wei Yang
UCSD: Matevz Tadel, Alja Mrak-Tadel
UNL: Brian Bockelman
WLCG: Fabrizio Furano, David Smith
US Department of Energy

Contract DE-AC02-76SF00515 with Stanford University
August 6-7, 2013
LSST Database Architecture Review
‹#›