The Fermilab Program 2005-2025 Michael Witherell Users meeting June 3, 2004.
Oct 24, 2002 Michael Ernst, Fermilab DRM for Tier1 and Tier2 centers Michael Ernst Fermilab February...
-
Upload
brianne-curtis -
Category
Documents
-
view
212 -
download
0
Transcript of Oct 24, 2002 Michael Ernst, Fermilab DRM for Tier1 and Tier2 centers Michael Ernst Fermilab February...
Oct 24, 2002Michael Ernst, Fermilab
DRM for Tier1 and Tier2 centers DRM for Tier1 and Tier2 centers
Michael ErnstMichael Ernst
FermilabFermilab
February 3, 2003February 3, 2003
February 3, 2003Michael Ernst, Fermilab 2
Production Data Flow Today …Production Data Flow Today …Tier1 at FNAL
dCache/Enstore
CATALOG
Local Disk
CATALOG
Tier2Worker Nodes Tier2
Head/Storage Node
CATALOG
MOP Master
February 3, 2003Michael Ernst, Fermilab 3
Typical Tier2 Today …Typical Tier2 Today …
Gri
dFT
P, b
bcp
Head/Storage Node
Worker Nodes
February 3, 2003Michael Ernst, Fermilab 4
Tier2 w/dCacheTier2 w/dCache
dCache
Admin Node
Gri
dF
TP
February 3, 2003Michael Ernst, Fermilab 5
Tier2 w/dCacheTier2 w/dCache
dCacheG
rid
FT
P
dCache
Gri
dF
TP
dCache/Enstore
Catalog
Gri
dF
TP
GridFTP
CASTOR
FNAL
CERN
UCSD
Florida, …
Cat
alog
Cat
alo g
Catalog
February 3, 2003Michael Ernst, Fermilab 6
dCache PlacementdCache Placement
Application (e.g. dccp)
dCap library
xxxFTP GRID Access Method(s)
Applications
dCache
Enstore
OSM
HSM X
Local Disk
PNFS NamespaceManager
Tertiary StorageSystems
February 3, 2003Michael Ernst, Fermilab 7
Distributed Pool ArchitectureDistributed Pool Architecture
Topic Caches
Tertiary Storage
Enstore, OSM, …
Super Cluster Caches(Experiment)
Central Cache
HostCaches
Cluster Caches(Working Group)
externally enforced attraction
destination determined attraction
February 3, 2003Michael Ernst, Fermilab 8
What is dCache ?What is dCache ?
dCachedCache is/provides is/provides• Compact system making access to HSM systems more efficientCompact system making access to HSM systems more efficient• Generic tool for caching, storing and easily accessing peta byte scale Generic tool for caching, storing and easily accessing peta byte scale
datasets distributed among a large set of heterogeneous caching nodesdatasets distributed among a large set of heterogeneous caching nodes• Able to run with and without HSM backend (can be used as scalable file Able to run with and without HSM backend (can be used as scalable file
store)store)• Very flexible HSM BackendVery flexible HSM Backend
Supports multiple instances of same HSM or different HSMs within same dCache Supports multiple instances of same HSM or different HSMs within same dCache instanceinstance
Rate Adaption (tape Rate Adaption (tape disk) disk) Deferred Write (aggregates small files up to threshold (time, space))Deferred Write (aggregates small files up to threshold (time, space)) StagingStaging Read AheadRead Ahead dCache allows to relax performance requirements on HSM storage components (e.g. dCache allows to relax performance requirements on HSM storage components (e.g.
tape drives and robots) w/o overall performance penaltytape drives and robots) w/o overall performance penalty
• Attraction schemes used to optimize data placementAttraction schemes used to optimize data placement• Load BalancingLoad Balancing• Automated Replication (to prevent hot spots)Automated Replication (to prevent hot spots)
February 3, 2003Michael Ernst, Fermilab 9
dCache Data Access dCache Data Access • LocalLocal
dccp (req. mounted pnfs fs), kerberized and non-kerberizeddccp (req. mounted pnfs fs), kerberized and non-kerberized dcap lib (API)dcap lib (API) dcap Preload lib dcap Preload lib URL-style Addressing (e.g. dccp URL-style Addressing (e.g. dccp
dcap://door.do.name:port#/pnfs/do.name/path/to/file)dcap://door.do.name:port#/pnfs/do.name/path/to/file) GridFTP (Server embedded in dCache, client with globus-url-copy)GridFTP (Server embedded in dCache, client with globus-url-copy)
• RemoteRemote URL-style AddressingURL-style Addressing GridFTPGridFTP
February 3, 2003Michael Ernst, Fermilab 10
The dCap LibraryThe dCap Library
Posix open/read/write/close dCache System
Door Node
Mover Node
Door Node
NFS Nodepnfsnative FS
libc.o
libdcache.so
Application
Data
Namespace Operations
February 3, 2003Michael Ernst, Fermilab 11
Current StatusCurrent StatusAccess to Mass Storage Systems (dCache/Enstore) at FNALAccess to Mass Storage Systems (dCache/Enstore) at FNAL• LocalLocal
dccp / encp (req. mounted pnfs fs)dccp / encp (req. mounted pnfs fs) URL-style AddressingURL-style Addressing GridFTP (Server embedded in dCache, client with globus-url-copy)GridFTP (Server embedded in dCache, client with globus-url-copy)
• RemoteRemote URL-style AddressingURL-style Addressing GridFTPGridFTP
in Service today atin Service today at– CERN (lxcmsa)CERN (lxcmsa)– UCSD (cms-dcache-serv, t2cms0)UCSD (cms-dcache-serv, t2cms0)– Using kerberized certificates when communicating w/dCache @ FNALUsing kerberized certificates when communicating w/dCache @ FNAL
» requires installation/configuration ofrequires installation/configuration of» NMI kx509/KCA bins,libsNMI kx509/KCA bins,libs» Kerberos bins, libs, krb5.conf Kerberos bins, libs, krb5.conf
• Mass Storage elsewhere Mass Storage elsewhere dCache up&running at UCSD dCache up&running at UCSD dCache @ CERN w/interface to CASTOR will be nextdCache @ CERN w/interface to CASTOR will be next
• Still MissingStill Missing Global Dataset Catalog, Replication ManagementGlobal Dataset Catalog, Replication Management
February 3, 2003Michael Ernst, Fermilab 12
Storage ManagementStorage Management• Evaluating Catalogs and Storage ManagersEvaluating Catalogs and Storage Managers
There is no LHC Data/Storage Management System, nor a Global Dataset Catalog as of yetThere is no LHC Data/Storage Management System, nor a Global Dataset Catalog as of yet US CMS is looking into SRB (developed by SDSC)US CMS is looking into SRB (developed by SDSC)
SRB servers are now running on US CMS servers at UCSD, Caltech, Fermilab, and at CERN SRB servers are now running on US CMS servers at UCSD, Caltech, Fermilab, and at CERN soonsoon
Distributed Storage Resources(database systems, archival storage systems, file systems, (Grid)FTP, http, …)
Application
SRB Server
MCAT
HRM DB2, Oracle, Illustra, ObjectStore HPSS, ADSM, dCache
• The Storage Resource Broker is a middleware• SRB is a Distributed Filesystem• It virtualises resource access• It mediates access to distributed heterogeneous resources• It uses a MetaCATalog to facilitate the brokering• It integrates data and metadata
February 3, 2003Michael Ernst, Fermilab 13
Global Data Management w/SRBGlobal Data Management w/SRB
…GridFTP
LocalMCAT
dCache
GridFTP
LocalMCAT
dCache
GridFTP
LocalMCAT
dCache
Global Catalog
Site A Site B Site N