Partner Logo Testbed 1 Software Ron Trompert ron

Partner Logo Testbed 1 Software Ron Trompert ron @ sara. nl Ron Trompert – Testbed 1 Software – 7 November 2001 - n° 1

Summary u Testbed 1 software presentation at CERN n Testbed overview n Job submission n User interface n Information system n WP 4 stuff n EDG Globus Configuration u Globus 2. 0 u SARA’s installation on a Compaq cluster contribution to testbed 1 Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 2

Testbed overview WP 2/Replica Catalogue II Input Sandbox UI JDL RB Input Sandbox Job Submit Event Output Sandbox L&B Brokerinfo JSS SE Output Sandbox Job Status CE Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 3

Job Submission # # ----- Job Description File ----# Executable = "WP 1 test. C"; Std. Input = "sim. dat"; Std. Output = "sim. out"; Std. Error = "sim. err"; Input. Sandbox = {"/home/wp 1/Hands. On-0409/WP 1 test. C", "/home/wp 1/Hands. On-0409/file*”, "/home/wp 1/DATA/*"}; Output. Sandbox = {"sim. err", "test. out", "sim. out"}; Rank = other. Average. SI 00; Requirements = (other. Op. Sys == "Linux RH 6. 1" || other. Op. Sys == "Linux RH 6. 2”) && (other. Run. Time. Environmnet == “CMS 3. 2”); Input. Data = "LF: test 10096 -0009"; Replica. Catalog = "ldap: //sunlab 2 g. cnaf. infn. it: 2010/rc=WP 2 INFN Test Replica Catalog, dc=sunlab 2 g, dc=cnaf, dc=infn, dc=it"; Data. Access. Protocol = "gridftp"; Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 4

User interface u dg-job-submit allows the user to submit a job for the execution on remote resources n n -r, -resource res_id the job is submitted by the Broker to the resource identified by res_id -i, -input_file the user must choose a resource id from a list of resources -n, -notify e_mail_address an e-mail message is sent to the specified e-mail address when the job enters in one of the following status: READY, RUNNING, ABORTED or DONE -o, -output out_file the generated dg_jobid is written in the file out_file Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 5

u dg-job-get-output requests to the Broker the job output files, specified by the Ouput. Sandbox attribute of the job-ad, and stores them on the submitting machine local disk u dg-job-list-match requirements returns the list of resources which fulfills job u dg-job-cancels one or more submitted jobs u dg-job-status displays bookkeeping information about submitted jobs u dg-job-get-logging-info submitted jobs displays logging information about Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 6

Information system u There n are two to pick from Globus MDS 2. 1 s Integrated in Globus s Memory caching, LDAPv 3, GSI authentication s Multiple VO’s on one node s GRIS and GIIS use the same slapd and listen to the same port. They are only distinguished through their DN’s: n n n GIIS: … Mds-Vo-Name=VO_NAME, o=Grid GRIS: … Mds-Vo-Name=local, o=Grid Open. LDAP Ftree s Runs independently of globus s Memory caching, LDAPv 3, GSI authentication s GRIS and GIIS on different ports Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 7

WP 4 stuff u Interim n Image Cloning s n Installation System Quick and easy installation of identical machines in large clusters Local Con. Fi. Guration system s s Handle automated installation and configuration in a very diverse and evolving environment Cfengine-like Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 8

Image cloning Configuration Management System Reference node Image server Boot server New node Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 9

LCFG configuration files mkxprof HTTP Web Server XML Profile (one per client node) rdxprof ldxprof LCFG SERVER Generic Component DBM File LCFG Components LCFG CLIENT Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 10

LCFG: RPM configuration u The list of RPMs to be installed on a node is defined in text files stored in the LCFG server u These RPM lists can be shared by several nodes; the LCFG configuration associates one list to each node u These files must be accessible to the clients using a remote file system (NFS) u The RPMs themselves must be also accessible via NFS u Any RPM installed manually in the client without being included in the files will be AUTOMATICALLY REMOVED Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 11

EDG Globus configuration u Everything u Security n n is installed in $GLOBUS_LOCATION i. e. /opt/globus files reside in /etc/grid-security CA certificates, CRLs and signing policy files in /etc/gridsecurity/certificates Grid-mapfile, host key and host certificate in /etc/grid-security u Globus setup packages are not used. They are replaced by the EDG config packages globus_<package>-edgconfig where <package> is the name of the globus setup package it replaces Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 12

u. A single configuration file /etc/globus. conf n n Local (internal) configuration files are (re)created by each restart of service with parameters from /etc/globus. conf Simple format: s # at start of line means a comment s MACRO=VALUE s Example: s GLOBUS_LOCATION=/opt/globus Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 13

Globus 2. 0 installation on a Compaq cluster u Build_from_source_tarballs n Perl script to build the distribution n Site specific information should be set: s my $SRC_TAR_LOC = “ … “; s my $SRC_LOC = “ …”; s my $GLOBUS_LOC = “ … “; s my $FLAVOR = “gcc 32 dbg”; n n Although the Compaq is a 64 -bit machine 32 -bit was specified here because when you use 64, the scripts think that you are on a SGI machine and invoke the MIPSpro --enable-64 bit compiler option. 32 does nothing. Haven’t tried vendorcc (ccc) yet s my $LOG_LOC = “ … “; s my $TMP_DIR = “ … “; Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 14

u Compilation n Globus-ssl-utils-2. 1 s Error message: undefined reference to ‘_Ots. Remainder 64 Unsigned’ Linking with –lots by editing the configure script Warning: gnu_regex. c: cast from pointer to integer of different size On the Compaq this did not seem to cause any problems. The machine only has 512 Mb of memory and tests showed that the high 32 bits of 64 -bit pointers did not change. n s n n Globus_gatekeeper. c & globus_gram_job_manager. c s Error: redefininition of sys_errlist[ ] For some reason “TARGET_ARCH_LINUX” was not #define-ed. A dirty hack solved the problem. n u Globus-script-pbs-submit n #PBS –l ncpus=$grami_count should be #PBS –l nodes=$grami_count n Set is_cluster=true Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 15

u Tell Globus where MPI is in globus-job-manager-tools. sh and globussh-tools. sh u LD_LIBRARY_PATH containing $GLOBUS_LOCATION/lib for globus-gatekeeper and in. ftpd n Solved by wrapper Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 16

SARA’s contribution to testbed 1 u Resources n Compute Elements s Compact 16 nodes 600 MHz alpha Myrinet 2000 interconnect Fast Ethernet interconnect Open. PBS Grid. FTP Jobmanagers fork and pbs MPI over myrinet and fast Ethernet. MPI over myrinet is the default, MPI over Fast Ethernet only through RSL. n n n n s Itanium cluster? Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 17

SARA’s contribution to the testbed 1 n Storage Element s Teras with Storage. Tek Near. Line storage n n 10 TB disk 120 TB tape Grid. FTP No jobmanagers Ron Trompert – Testbed 1 Software- 7 November 2001 - n° 18