special jobs

www.eu-eela.org

E-infrastructure shared between Europe and Latin America

FP6−2004−Infrastructures−6-SSA-026409

Special Jobs

Valeria Ardizzone

INFN - Catania

12th EELA Tutorial

Lima, 26.09.2007

2


Lima, 12th EELA Tutorial, 25.09.2007FP6−2004−Infrastructures−6-SSA-026409

Outline• Overview Job with Data Requirements

• Overview MPI

- How to create a MPI job.

- MPI job in middleware.

• Overview DAG

Outline



• User can stage input files from the UI to the WN using the InputSandbox attribute

InputSandbox = {"codesa.i686",

"start_root.sh",

"./Korba/atmbc.const",

"./Korba/bctran-window.3",

"./Korba/codesa3d.fnames",

".rootrc",

"convert.C",

"GraphCODESA3D.C"};

Overview Jobs with Data

The upper limit for

InputSandbox is 10Mbyte!

4



What can I do if my job requires huge data to be processed ?

..the question

5



InputDataInputData (optional)

This is a string or a list of strings representing the Logical File Name (LFN) or Grid Unique Identifier (GUID) needed by the job as input.

The list is used by the RB to find the CE from which the specified files can be better accessed and schedules the job to run there.

InputData = {“lfn:cmstestfile”,InputData = {“lfn:cmstestfile”,

““guid:135b7b23-4a6a-11d7-87e7-9d101f8c8b70”};guid:135b7b23-4a6a-11d7-87e7-9d101f8c8b70”};

..the answer /1

6



DataAccessProtocolDataAccessProtocol (mandatory if InputData has been specified)

The protocol or the list of protocols which the application is able to “speak” with for accessing files listed in InputData on a given SE.

Supported protocols in gLite are currently gsiftpgsiftp, and filefile.

DataAccessProtocol = {“file”,“gsiftp”};DataAccessProtocol = {“file”,“gsiftp”};

..the answer /2

7



pds2jpg-MERIS-Etna.jdl

[ JobType = "normal"; Type = "Job";

Executable = "/bin/bash"; Arguments = "pds2jpg_install.sh \ MER_FR__2PNUPA20030504_092534_000000502016_00079_06145_0033";

StdOutput = "pds2jpg.out"; StdError = "pds2jpg.err";

InputSandbox = {"./pds2jpg_install.sh","./beam20.tar.gz"};

InputData = {"lfn:/grid/gilda/MER_FR__2PNUPA20030504_092534_000000502016_00079_06145_0033.N1"}; DataAccessProtocol = {"gridftp","rfio","gsiftp"};

OutputSandbox = { "MER_FR__2PNUPA20030504_092534_000000502016_00079_06145_0033.jpg", "ENVISAT_Product_courtesy_of_European_Space_Agency", "pds2jpg.out", "pds2jpg.err" };

RetryCount = 3; ]

8



pds2jpg_install.sh

#!/bin/shecho Staging Input Data $Courtesy of European Space Agency$;#Skip the "/" from the argument.file=`echo $1 | awk -F '/' '{print $2}'`echo lcg-cp --vo gilda lfn:/grid/gilda/MER_FR__2PNUPA20030504_092534_000000502016_00079_06145_0033.N1file:`pwd`/${file}.N1lcg-cp --vo gilda lfn:/grid/gilda/MER_FR__2PNUPA20030504_092534_000000502016_00079_06145_0033.N1 file:`pwd`/${file}.N1echo Staging Application;ls -algunzip beam20.tar.gz;tar xvf beam20.tar;cd beam-2.0/bin;echo Starting Application;echo "./pds2jpg-run.sh $file;"./pds2jpg-run.sh $file;echo "mv $file.jpg ../.."mv $file.jpg ../..touch ../../ENVISAT_Product_courtesy_of_European_Space_Agencyecho "Input ENVISAT Product courtesy of European Space Agency">../../ENVISAT_Product_courtesy_of_European_Space_Agencyecho No Output Packaging;

9



the output..

Input ENVISAT Product

courtesy of European

Space Agency

10



..another couple of questions

The output produced by my job must be processed, as input, by some other jobs

Q.1) How can I make accessible this data for other computation ?

Q.2) Can I upload the data to be processed later automatically ?

11



OutputDataOutputData (optional)

This attribute allows the user to ask for the automatic upload and registration of datasets produced by the job on the Worker Node (WN).

This attribute contains the following three attributes:

1. OutputFile

2. StorageElement

3. LogicalFileName

Output data

12



OutputFileOutputFile (mandatory if OutputData has been specified)This is a string attribute representing the name of the

output file, generated by the job on the WN, which has to be automatically uploaded and registered by the WMS.

StorageElementStorageElement (optional)This is a string representing the URI of the Storage

Element where the output file specified in the OutputFile has to be uploaded by the WMS.

LogicalFileNameLogicalFileName (optional)This is a string representing the LFN user wants to

associate to the output file when registering it to the Catalogue.

Automatic uploading mechanism

NOT yet supported in gLite

Output data

13



• Execution of parallel jobs is an essential

issue for modern conceptions of

informatics and applications.

• Most used library for parallel jobs support is (Message Passing Interface) MPI

• At the state of the art, parallel jobs can run inside single Computing Elements (CE) only;

– several projects are involved into studies concerning the possibility of executing parallel jos on Worker Nodes (WNs) belonging to differents CEs.

Overview MPI

14



How to create a MPI Job

15



• In order to garantee that MPI job can run, the following requirements MUST BE satisfied:

– the MPICH MPICH software must be installed and placed in the PATH environment variable, on all the WNs of the CE.

– The Executable that is specified in the JDL must not be the MPI application directly, but a wrapper script that invokes the MPI applications by calling mpirun command.

Requirements & Settings

16



• For the user’s point of view, jobs to be run as MPI are specified setting the JDL JobType attribute to MPICH and specifying the NodeNumber attribute as well.

E.g.:

JobType = “MPICH”;

NodeNumber = 4; This attribute define the required number of CPUs needed for the application.

17



• When these two attributes are included in a JDL the User Interface (UI) automatically add the following expression

(other.GlueCEInfoTotalCPUs >= NodeNumber) &&

Member (“MPICH”,other.GlueHostApplicationSoftwareRunTimeEnvironment)

to the JDL requirements expression in order to find out the appropriate resources where the job can be executed.

18



the problem ...

• Unfortunately LCG project was not

synchronized with the latter requirement

avoiding to share disk space with nodes

inside the same CE.

• This drove us to spend our time in providing

a ad-hoc solution in order to find an efficent workaround to this problem.

• The solution adopted bypasses the problem by putting some intelligence inside the script passed in Inputsandbox.

19



… the solution

• In detail each job has to mirror, via scp, its files on all nodes dedicated to it.

ssh hostbased authentication MUST BE well configured between all the WNs.

20



[

Type = "Job";

JobType = "MPICH";

Executable = "MPItest.sh";

NodeNumber = 5;

Arguments = "cpi 5";

StdOutput = "test.out"; StdError = "test.err";

InputSandbox = {"MPItest.sh","cpi"};

OutputSandbox = {"test.err","test.out","executable.out"};

Requirements = other.GlueCEInfoLRMSType == "PBS" || other.GlueCEInfoLRMSType == "LSF";

]

mpi.jdl

Actually the Local ResourceManager

supported are PBS and LSF only.

Actually the Local ResourceManager

supported are PBS and LSF only.

The number of threads specified with NodeNumber attribute agrees with the second Argument. It will be used during the invoking of mpirun command.

The number of threads specified with NodeNumber attribute agrees with the second Argument. It will be used during the invoking of mpirun command.

21



for i in `cat $HOST_NODEFILE$HOST_NODEFILE` ; doecho "Mirroring via SSH to $i"# creates the working directories on all the nodes allocated for parallel execution.ssh $i mkdir -p `pwd`# copies the needed files on all the nodes allocated for parallel execution./usr/bin/scp -rp ./* $i:`pwd`# checks that all files are present on all the nodes allocated for parallel execution.ssh $i ls `pwd`

done

# execute the parallel job with mpirun.echo "Executing $EXE"chmod 755 $EXE

mpirun -np $CPU_NEEDED -machinefile $HOST_NODEFILE `pwd`/$EXE > executable.out

MPItest.sh

The Environment variable $HOST_NODEFILE contains the list of WNs allocated for the parallel execution.

The Environment variable $HOST_NODEFILE contains the list of WNs allocated for the parallel execution.

22



DAG job

• A DAG job is a set of jobs where input, output, or execution of one or more jobs can depend on other jobs

• Dependencies are represented through Directed Acyclic Graphs, where the nodes are jobs, and the edges identify the dependencies

nodeA

nodeB nodeC NodeF

nodeD

23



JDL structure

24



Attribute: Nodes

25



Attribute: Dependencies

26



DAG jdl

[ type = "dag"; max_nodes_running = 4; nodes = [ nodeA = [ file ="nodes/nodeA.jdl" ; ]; nodeB = [ file ="nodes/nodeB.jdl" ; ]; nodeC = [ file ="nodes/nodeC.jdl" ; ]; nodeD = [ file ="nodes/nodeD.jdl"; ]; dependencies = { {nodeA, nodeB}, {nodeA, nodeC}, { {nodeB,nodeC}, nodeD } } ];]

Node description could also be done here,

instead of using separate files

27



References

• Job Description Language https://edms.cern.ch/file/555796/1/EGEE-JRA1-TEC-555796-JDL-Attributes-v0-8.pdf

• GILDA wiki:– Job with Data https://grid.ct.infn.it/twiki/bin/view/GILDA/JobData

– MPI Job

https://grid.ct.infn.it/twiki/bin/view/GILDA/MPIJobs-withedgcommands

29



Thank you very much for your kind

attention!

special jobs

Documents