Revision b08ff89cf10a00cdf4d7c9815b2b732c986934ec authored by Fons Rademakers on 12 March 2004, 00:31:22 UTC, committed by Fons Rademakers on 12 March 2004, 00:31:22 UTC
git-svn-id: http://root.cern.ch/svn/root/trunk@8372 27541ba8-7e3a-0410-8455-c3a389f83636
1 parent 758a188
README.PROOF
Installing and using the Parallel ROOT Facility -- PROOF
========================================================
PROOF allows the parallel execution of ROOT scripts and the
parallel analysis of ROOT trees on a heterogeneous cluster.
To install and use PROOF on a cluster do the following:
0) Make sure each node in the cluster has /usr/local/root
or /opt/root. This can be directly an un-tared binary
distribution of ROOT or a link to a centrally installed
copy of ROOT.
1) Add to /etc/services the lines:
proofd 1093/tcp
rootd 1094/tcp
2) Add to /etc/inetd.conf the lines:
proofd stream tcp nowait root /usr/local/root/bin/proofd proofd -i /usr/local/root
rootd stream tcp nowait root /usr/local/root/bin/rootd rootd -i
Or when using xinetd add the following two files to /etc/xinetd.d:
# file name: rootd
# default: on
# description: The rootd daemon allows remote access to ROOT files.
service rootd
{
disable = no
socket_type = stream
wait = no
user = root
server = /usr/local/root/bin/rootd
server_args = -i /usr/local/root
log_on_success += DURATION USERID
log_on_failure += USERID
}
# file name: proofd
# default: on
# description: The proofd daemon is used to startup PROOF.
service proofd
{
disable = no
socket_type = stream
wait = no
user = root
server = /usr/local/root/bin/proofd
server_args = -i /usr/local/root
log_on_success += DURATION USERID
log_on_failure += USERID
}
In above examples correct daemon path to reflect your setup.
Restart (x)inetd or force it to re-read the config file
(kill -1 <inetd.pid>).
3) (Optional) add to /etc/syslog.conf on the master node the line:
local5,local6.debug /usr/local/root/proof/log/proof.log
and all slave nodes:
local5,local6.debug @master.cern.ch
where master.cern.ch is domain name of the master node.
All PROOF syslog messages will be collected on the master node.
Just make one node in the cluster the master, all others are slaves.
And change on all nodes the line:
*.info;mail.none;news.none;authpriv.none /var/log/messages
to:
*.info;local5,local6,mail.none;news.none;authpriv.none /var/log/messages
Create an empty /usr/local/root/proof/log/proof.log.
Restart syslogd or force it to re-read the config file
(kill -1 <syslogd.pid>).
4) Edit /usr/local/root/proof/etc/proof.conf to reflect your cluster
configuration. See the example proof.conf file for more information.
5) This is all. To test PROOF try the following:
% root
root [1] gROOT->Proof("<master.node.ch>")
root [2] gProof->Print()
<shows information on the master and all active slave servers>
root [3] .q
6) (Optional) Edit /usr/local/root/proof/etc/motd (Message of the day)
The contents of this file will be displayed in the client when a
PROOF session is started.
7) (Optional) To disable the use of the cluster temporarily you can
create the file /usr/local/root/proof/etc/noproof. When a user
tries to start a PROOF session the contents of the file will be
displayed and no session will be started.
==================================
The /usr/local/root/proof/utils directory contains some useful
shell scripts for managing your cluster. They use the file
/usr/local/root/proof/etc/cluster.conf to define the hosts they
operate on.
(As an alternative, some people are happy using dsh (distributed
shell) from http://dsh.sourceforge.net)
Files distributed with /usr/local/root/proof/utils/push:
/etc/passwd
/etc/group
/etc/exports
/etc/services
/etc/inetd.conf
Computing file changes ...