next up previous contents index
Next: KCAP Web-based Cluster Management Up: SCE: Scalable Cluster Environment Previous: Building Cluster with Beowulf   Contents   Index

Subsections


SCMS Cluster Management System


Overview

SCMS - Scalable Cluster Management System - is a very interactive and extensible cluster management tool. The goal of SCMS is to allow users to perform the administrative task easily. SCMS provides powerful graphics user interface, huge amount of command tools, real-time monitoring subsystem, web-based interface and more. With SCMS, the system administration task of large cluster will become much simpler.


Features


Requirements


Installation

  1. Extract the archive.
    
    # tar xzvf scms-2.0.tar.gz
    

  2. Run ./configure to configure the system
    
    # cd scms-2.0
    # ./configure
    
    SCMS fully supports GNU-style autoconf which provides additional arguments listed below:

  3. Run make to compile.
    
    # make
    

  4. Run make install to install.
    
    # make install
    

  5. Run make rpm to build RPM.
    
    # make rpm
    


Manual


Running

To start scms, simply type scms


$ scms


Configuration

SCMS configuration files are located in ETCDIR/sce, there are 2 necessary files:

  1. scms
  2. sce.conf


ETCDIR/sce/scms

scms uses shell-style configuration file: Sharp sign (#) at the first character indicates commentation; An assignment forms by key and value seperated by equal sign (=) without among additional spaces. This file must be duplicated to every nodes in the cluster.



#
# scms startup configuration
#

# hostname or ip which you supposed to run configuration server
CONFIGD_ADDR=amata1

# absolute path to configuration file
CONFIG_FILE=/etc/sce/sce.conf


ETCDIR/sce/sce.conf

sce.conf uses section/option-style based on ConfigParser in python. There are few necessary sections:

  1. main - For future using.
  2. scms - Keep SCMS configuration, specially network ports.
  3. plugin - Specify default plug-in.
  4. hostlist - List all host's name.
  5. partlist - List all partition's name.
  6. hostpart - Associate host to partition.
  7. part* - Subsection of each partition represents host in partition and SMA address for them.



[main]

[scms]
sma_port=6011
cma_port=6012
rmi_port=6013
sma_rlog_port=9900
cma_rlog_port=9910

[plugin]
#plugin_path=<plugin_path>
plugin_num=2
0=libhwplugin.so
1=libpsplugin.so

[hostlist]
host_num=12
0=amata1
1=amata2
2=amata3
3=amata4
4=amata5
5=amata6
6=amata7
7=amata8
8=amata9
9=amata10
10=amata11
11=amata12

[partlist]
part_num=1
0=part1

[hostpart]
158.108.35.35=part1
158.108.35.36=part1
158.108.35.37=part1
158.108.35.38=part1
158.108.35.39=part1
158.108.35.40=part1
158.108.35.41=part1
158.108.35.42=part1
158.108.35.43=part1
158.108.35.44=part1
158.108.35.45=part1
158.108.35.46=part1
158.108.35.35_s=part1

[part1]
sma_addr=amata1
host_num=12
0=158.108.35.35
1=158.108.35.36
2=158.108.35.37
3=158.108.35.38
4=158.108.35.39
5=158.108.35.40
6=158.108.35.41
7=158.108.35.42
8=158.108.35.43
9=158.108.35.44
10=158.108.35.45
11=158.108.35.46


Real-time Monitoring Subsystem

SCMS/RMS consists of 3 necessary daemons:

  1. System Management Agent (SMA)
  2. Control and Monitoring Agent (CMA)
  3. Resource Management Agent (RMI)

Normally, CMA should always run on all nodes and periodically forwords data to its SMA. Using above method, SCMS allows administrator to configure flexible connection in hierarchical structure by setting SMA to have its SMA.

We define Partition as a group of CMAs or SMAs and a SMA. That means each CMA or SMA in the same partition have the same SMA.

After installation, all three daemons are started as system services at start up (RedHat). The user may start it manually using init.d script.


# /etc/rc.d/init.d/configd start
# /etc/rc.d/init.d/sma start
# /etc/rc.d/init.d/cma start

Note that configd must be started before sma and cma. However, in case of the user want to stop/start/restart SCMS from scratch, SCMS provides the special scripts:


# cms_start
# cms_stop

next up previous contents index
Next: KCAP Web-based Cluster Management Up: SCE: Scalable Cluster Environment Previous: Building Cluster with Beowulf   Contents   Index
Sugree Phatanapherom
2001-06-21
I also have a line of punk t-shirts and art t-shirts featuring Bas Couture, artcore designs