srm & se jens g jensen wp5 atf, december 2002. collaborators rutherford appleton (atlas...

17
SRM & SE Jens G Jensen WP5 ATF, December 2002

Upload: leonard-randall

Post on 13-Dec-2015

216 views

Category:

Documents


1 download

TRANSCRIPT

Page 1: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

SRM & SE

Jens G Jensen

WP5

ATF, December 2002

Page 2: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

Collaborators

• Rutherford Appleton (ATLAS datastore)

• CERN (CASTOR)

• Fermilab

• Jefferson Lab

• Lawrence Berkeley– Americans have: ENSTORE, SAM, …

• Also present: WP2

For an overview of storage systems see D5.1

Page 3: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

File pinning & replication

LocalSRM

RemoteSRM

RC

filefile(replica)

• Client fetches a file using srmGet(LFN,SURL)• Local SRM gets a replica identified by LFN.• Issue: LFN is optional

Other issues:

What are defautl ACLs on replicas?

Does the local SRM have to check ACLs from the remote SRM?

Page 4: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

RCs• SRM must be able to interoperate with

several different RCs

• Getting SURLs (RC SRM) is easy, but:– Does SRM need reverse lookup (SURL

LFN)?

– How to notify RC that a file has been deleted?• Add hooks to SRM?

Page 5: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

SRM file lifetime overviewFiles are:• Volatile (e.g. pinned files)

– Files get deleted by SE when they time out– Files created as volatile are never written to tape

• Durable (e.g. permanent but not in final location)– Files can be deleted only by administrator or owner– Files still have a lifetime – owner and/or admin

notified when lifetime expires

• Permanent (e.g. your important experiment)– Files can be deleted only by administrator or owner

Page 6: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

New lifetime stuff: space

• Each user has space with a per-user (probably) quota for each type of space

• Users must reserve space before they can reserve for writing or reading

• Users are charged for the type and amount of space they reserve

• Space can be– Volatile – reservations not guaranteed beyond a minimum– Durable – notify user if file didn’t move, space wasn’t

released, etc – if user doesn’t react, notify admin– Permanent

• SRM Proposal: space lifetime is encoded in the filepath!!! (JLAB has this)

Proposal: space has a lifetime as well.

Page 7: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

Storage Semantics

Volatile Durable Permanent

Page 8: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

copyToSpace

• Users can “give away” files in their own space to another user– First step: the user who owns the files gives them

away– Second step: the other user takes over the files– Done without copying in the SRM, but quotas are

adjusted accordingly

• Alternatively, let users “give” files to other users without deleting them from their own space– copy-on-write

Page 9: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

SRM possibly wildly inaccurate status

GetPULL

PutPUSH

CopyFile

Lifetime

Space

LifetimeGSIFTP

SESE 1.0

TB 2.0

SE 1.0

TB 2.0

SE 1.1

TB 2.1

SE 1.1

TB 2.1

SE 1.x

x>=1

SE 1.0

TB 2.0

Fermi OK OK OK OKonly vol

-- OK

JLAB OK OK OK OK -- OK

Page 10: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

Comparing SRM v2.{0,1} to SE 1.0

• Get, Put: supported– Get in PULL mode (same as SRM)– Put in PUSH mode (same as SRM)

• File lifetime: will be easy to support– Currently (pre-1.0) all files are permanent…

• Space management – currently supported as volatile, best-effort only (i.e. very little support)

• Information providers: supported soon• Directory stuff : mostly supported (not rmdir, no

recursion)• Support for arbitrarily long LFNs: OK

Need to have API for more detailed comparison…

Page 11: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

Need for POSIX style access

• RFIO (CERN, and other sites with HPSS)

• DCAP (FermiLab)

• Globus XIO?

• file:

Pick one, and call it a standard?

Page 12: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

Security• Americans expect to use CAS

– Possibly need for interim solution until CAS is sufficiently mature?

• Need for “weak” ftp?– Users “delegate” username/password to SRM

• Europeans expect to use VOMS– SE solution until VOMS is sufficiently mature:

GACL (SE will always support GACL)– SE can use CAS as well but support not planned

• Use GSI-HTTPS for now – common ground essential for interoperability

Question: are CAS and VOMS interchangeable?

Page 13: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

ACLs• How to manage ACLs on replicas?

– Replicas are read-only; or– Replicas are writable by whoever created the replica– SRM may have to contact the SRM containing the

master to pick up changes to the ACL

• SRM proposal: default is world readable for volatile space, owner access only for permanent

• SE proposal– Have default ACL per user and/or per directory

Page 14: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

Web services

• Fermi uses GLUE but plans to switch to Apache/Tomcat/Axis.

• “Secret” SRM v1.0 WSDL available…

• New WSDL spec for v2.1 specification in preparation (Fermi)

• SOAP interoperability issues, particularly with parameters more complicated than strings and ints – e.g. (WSDL) structs.

• Need to be OGSI compliant…

Page 15: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

Schema

• Agree that GLUE is important

• More effort needed?

Page 16: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

How to ensure that users behave?

• Good behaviour == close file && release pin• Make them pay… Charge per space reservations

and/or size of files in user’s reservation?– CASTOR charges per MB on tape

• Count number of expired pins (in durable and permanent only!?)

• What to do with files that are opened and never closed?– Can be detected sometimes (local files, NFS?,

/grid?)– CASTOR currently does nothing

Page 17: SRM & SE Jens G Jensen WP5 ATF, December 2002. Collaborators Rutherford Appleton (ATLAS datastore) CERN (CASTOR) Fermilab Jefferson Lab Lawrence Berkeley

Collaborations…

• Agree standard protocols, accept each others certificates, etc.

Testing Interoperability

• Sharing code is seen as less important– Occasional licence conflict– People would rather have more independent

implementations