Install Pre-Req

We assume that you have already following the instructions here for installing GridFTP on top of HDFS.  Let $VDT_LOCATION be the location of your VDT install.  Initialize your environment:

source $VDT_LOCATION/setup.sh
source /opt/pacman/pacman-3.26/setup.sh

Replace the above lines with those for your site.

Install Hadoop, FUSE, and a GridFTP server

We need a valid install of Hadoop, mounted through FUSE.  A GridFTP-HDFS server must also be installed, but this does not need to be on the same server as the BestMan server.  The installation instructions are:

• Hadoop and FUSE.
• GridFTP-HDFS.
  

A larger install will prefer to have their GridFTP and BeStMan servers installed on separate hosts.

Install BeStMan

Install the latest BeStMan from the OSG with pacman:

pacman -get http://t2.unl.edu/store/cache:Hadoop-Bestman

If you are using GUMS, make sure you have set VDT_GUMS_HOST to the hostname of the GUMS server.  This will install the BeStMan server with the Ganglia load-balancing plugin.

Configure BeStMan

The BeStMan configuration file is in $VDT_LOCATION/bestman/conf/bestman.rc.  You'll probably want to take a look over that file.  The following configurations are interesting:

• supportedProtocolList: This should be a semi-colon separated list of accessible servers; should include the protocol, hostname, and port.  Example: gsiftp://red-gridftp1.unl.edu:5000;gsiftp://dcache-s01.unl.edu:5000
  
• noSudoOnLs (default: True): This defaults to True; set to False if you want BeStMan to use sudo to perform ls.  Use this if user daemon can't list a portion of your namespace.
  
• staticTokenList: A list of staticly-configured space tokens.  See the in-file help text for this information.
  
• GUMSserviceURL: Check this to make sure it points to the correct GridFTP.
• accessFileSysViaSudo: Set this to true or all your srmMkdir commands will result in directories owned by root.
  

Add the following lines to /etc/sudoers so that BeStMan can manipulate the filesystem namespace.

    Cmnd_Alias SRM_CMD = /bin/rm, /bin/mkdir, /bin/rmdir, /bin/mv, /bin/cp, /bin/ls
    Runas_Alias SRM_USR = ALL, !root
    daemon     ALL=(SRM_USR) NOPASSWD: SRM_CMD

    # NOTE: on RHEL 5+ systems you need to make sure the following is commented out
    #Defaults    requiretty

Make sure you actually commented out the "requiretty" line if it is present on your system.

Customizations for BeStMan

At Nebraska, we have the following customizations:

• JMX monitoring.  This allows us to closely monitor memory usage and thread activity in a much more Java-centric manner than Ganglia monitoring.
• To enable this, we add the following command-line arguments to the call to java in $VDT_LOCATION/bestman/sbin/bestman.server:
  

  -Dcom.sun.management.jmxremote.authenticate=false -Dcom.sun.management.jmxremote=true -Dcom.sun.management.jmxremote.ssl=false -Dcom.sun.management.jmxremote.port=8004

  Note that this allows anyone with access to port 8004 access to read out your JVM internal metrics.
  
• Ganglia-based selection of GridFTP servers.  The default algorithm for GridFTP server selection is round-robin through a static list.  We have a protocol plugin for BeStMan that randomly selects the servers from the list with a probability distribution function built from the load and memory usage as reported by Ganglia.  To enable this, add the following line to $VDT_LOCATION/bestman/conf/bestman.rc:
  

  protocolSelectionPolicy=class=edu.unl.rcf.BestmanGridftpSelector.BestmanGridftp&jarFile=UNLGangliaBestman.jar&name=gsiftp

  Then, add the following command-line arguments to the call to java in $VDT_LOCATION/bestman/sbin/bestman.server:
  

  -Dedu.unl.rcf.BestmanGridftpSelector.host=localhost -Dedu.unl.rcf.BestmanGridftpSelector.port=8649

  Add this after the ${MAXHEAP} parameter.