DRAFT version:  This page compiled by Eric Hjort, ELHjort@lbl.gov. 
DRM Testing Summary

VDT DRM binary is DRM v1.2 obtained fromVDT v1.2.2.

SDM RH8 binary is obtained from the SDM group's  download page.

Overview:  I did a non-root installation of VDT according to the instruction on the VDT page.  Initially I ran into some problems when I tried to do my own pacman installation, however, when I followed the pacman installation instructions on the VDT page all worked well.  Once I had pacman installed the VDT installation worked without complication.  DRM is included in VDT as an add-on and is installed seperately with pacman.  This also worked fine.  What you get in the VDT DRM is a binary DRM executable that is RH7 compatible.  I tested it on pdsfgrid4 and I didn't run into any problems.  Binaries are also available for other system versions from the SDM group's download page.  I tested the RH8 binary successfully on stargrid03 but it did not work on pdsfgrid4 (running RH8 via chos).  In the case that the binary does not work it is then necessary to build DRM from the source files.  This results in executables that function properly.  These results are summarized in the table below:

NODE
 VDT DRM binary (RH7)
 SDM RH8 DRM binary
 RH8 build from sources
pdsfgrid4.nersc.gov


stargrid03.rcf.bnl.gov
 N/A


Examples of STAR HRM Data Transfer Modes (BNL -> LBNL)
In STAR HRMs have been used for production data transfer from BNL (RCF) to LBNL (NERSC/PDSF) since 2002.  Described below are the two modes of HRM transfers (get files from NFS disks or HPSS) that are used for essentially all of our transfers.
We almost always use the srm-copy option -f which means that a list formatted as <sourceURL> <size> <targetURL> is read in from a file.  One reason  this is used is because it allows the targetURL to be specified for each file.  An example of such an srm-copy command would look like this:

srm-copy.linux -d -f productionHighReversedFullFieldP04ik.rndm -c /auto/u/hjort/hrm2/hrm.rc -l P04ik.log
This command would be run at the destination of the transfers (pdsfgrid4 in this example) since SRM transfers always work in "pull" mode.  Each line in the file productionHighReversedFullFieldP04ik.rndm looks like this:

gsiftp://stargrid03.rhic.bnl.gov/star/data32/reco/productionHigh/ReversedFullField/P04ik/2004/067/st_physics_5067059_raw_2030001.MuDst.root 278979705 srm://garchive.nersc.gov/nersc/projects/starofl/reco/productionHigh/ReversedFullField/P04ik/2004/067/st_physics_5067059_raw_2030001.MuDst.root

This file is generated by a STAR-specific script (diskOrHPSS.pl) that compares the RCF (mirror) and PDSF file catalogs.  Here the files are sourced from NFS disks at RCF instead of from HPSS.  This has proven to be a dependable means of transferring files and usually provides better throughput than if files are sourced from HPSS.  Note that this method doesn't even require an HRM to be running at RCF.  An HRM is running at PDSF which caches the files prior to sinking them into HPSS.  Once the files are in HPSS RRS is called and the files are entered into the PDSF file catalog.  The .rndm file extension denotes that the script has randomized the order of files in order to distribute the i/o load on the NFS disks.

Files that are not available on NFS disks can be sourced from HPSS at RCF.  The syntax used for these transfers is similar to that shown above:

srm-copy.linux -d -f productionLowFullFieldP04ik.srm -c /auto/u/hjort/hrm2/hrm.rc -l P04ik.log -at PLAIN -et GSI -al starpftp -ap "password"
and the file list for transfers from HPSS looks like this:

srm://stargrid03.rhic.bnl.gov:NSPORT/home/starreco/reco/productionHalfHigh/HalfField/P04ik/2004/059/st_physics_5059055_raw_1020009.MuDst.root?remoteobj=HRMServerBNL&msshost=hpss.rcf.bnl.gov&mssport=MSSPORT 268435346 srm://garchive.nersc.gov/nersc/projects/starofl/reco/productionHalfHigh/HalfField/P04ik/2004/059/st_physics_5059055_raw_1020009.MuDst.root

Here "NSPORT" is the stargrid03 port that the nameserver is running on and "MSSPORT" is the port used to access HPSS.  The source file now begins with srm: instead of gsiftp: and this mode of transfers requires HRMs to be running both at RCF and PDSF.

In practice it works best to break up large transfers (>10k files) into smaller transfers (<10k files).  A simple shell script is generated by the diskOrHPSS.pl script to do a series of srm-copy command, each command corresponding to a particular trgsetupname/magscale/production of STAR MuDst's:

#! /usr/local/bin/tcsh
srm-copy.linux -d -f productionHalfHighHalfFieldP04ik.rndm -c /auto/u/hjort/hrm2/hrm.rc -l P04ik.log
srm-copy.linux -d -f productionHalfLowHalfFieldP04ik.rndm -c /auto/u/hjort/hrm2/hrm.rc -l P04ik.log
srm-copy.linux -d -f productionPPReversedFullFieldP04ik.rndm -c /auto/u/hjort/hrm2/hrm.rc -l P04ik.log

Note that at present it is not possible to combine files in HPSS and on NFS disk into a single file list.  This is because the authentication methods for disk (GSI, the default) and HPSS (PLAIN) are different.  Hopefully we'll get GSI authentication at RCF HPSS working in the near future.

For more information about SRM data transfers see the DataMover-UserGuide which is also linked from the SDM group's webpages.