1. Introduction

Ganglia provides a complete real-time monitoring and execution environment that is in use by hundreds of universities, private and government laboratories and commercial cluster implementors around the world. Ganglia is as simple to install and use on a 16-node cluster as it is to use on a 512-node cluster as has been proven by its use on multiple 500+ node clusters.

Ganglia was initially developed at the University of California, Berkeley Computer Science Division as way to link clusters across the Berkeley campus together in a logical way. Since it was developed at a university, it is completely open-source and has no proprietary components. All data is exchanged in well-defined XML and XDR to ensure maximum extensibility and portability.

The monitoring core allows you to monitor any number of host metrics in real-time. At present, the monitoring core runs on Linux, FreeBSD, Solaris, AIX, Tru64 and IRIX. There is a Windows port as well which is in beta.

Ganglia is not just a way to link nodes in a cluster together in a logical way but also a way to link clusters to other clusters. Ganglia blurs the line between clustering and distributed computing by providing for Cluster to Cluster (C2C) data exchanges which link disparate cluster resources together into a single logical framework.

telnet localhost 8649

2. Installation

3. Configuration

---

3.1. Gmetad

The behavior of the Ganglia Meta Daemon is completely controlled by a single configuration file which is by default /etc/gmetad.conf. For gmetad to do anything useful you much specify at least one data_source in the configuration. The format of the data_source line(s) is as follows

data_source "Cluster A" 127.0.0.1 1.2.3.4:8655 1.2.3.5:8625 data_source "Cluster B" 1.2.4.4:8655

In this example, there are two unique data sources: "Cluster A" and "Cluster B". The Cluster A data source has three redundant sources. If gmetad cannot pull the data from the first source, it will continue trying the other sources in order.

If you do not specify a port for a source, gmetad will assume you want to connect to the default gmond port of 8649.

Here is a sample gmetad configuration file:

# This is an example of a Ganglia Meta Daemon configuration file # http://ganglia.sourceforge.net/ # # $Id: gmetad.conf,v 1.3 2002/09/19 18:56:42 sacerdoti Exp $ # # Setting the debug_level above zero will make gmetad output # debugging information and stay in the foreground # default: 0 # debug_level 10 # # The data_source tag must immediately be followed by a unique # string which identifies the source then a list of machines # which service the data source in the format ip:port, or name:port. # If a # port is not specified then 8649 (the default gmond port) is # assumed. # default: There is no default value # data_source "my box" localhost my.machine.edu:8655 1.2.3.5:8655 # data_source "another source" 1.3.4.7:8655 1.3.4.8 # # List of machines this gmetad will share XML with # default: There is no default value # trusted_hosts 127.0.0.1 169.229.50.165 # # If you don't want gmetad to setuid then set this to off # default: on # setuid off # # User gmetad will setuid to (defaults to "nobody") # default: "nobody" # setuid_username "nobody" # # The port gmetad will answer requests for XML # default: 8651 # xml_port 8651 # # The number of threads answering XML requests # default: 2 # server_threads 4 # # Where gmetad stores its round-robin databases # default: "/var/lib/ganglia/rrds" # rrd_rootdir "/some/other/place"

# $Id: gmond.conf,v 1.2 2002/09/19 00:37:18 sacerdoti Exp $
# This is the configuration file for the Ganglia Monitor Daemon (gmond)
# Documentation can be found at http://ganglia.sourceforge.net/docs/
#
# To change a value from it's default simply uncomment the line
# and alter the value
#####################
#
# The name of the cluster this node is a part of
# default: "unspecified"
# name  "My Cluster"
#
# The owner of this cluster. Represents an administrative
# domain. The pair name/owner should be unique for all clusters
# in the world.
# default: "unspecified"
# owner "My Organization"
#
# The latitude and longitude GPS coordinates of this cluster on earth.
# Specified to 1 mile accuracy with two decimal places per axis in Decimal
# DMS format: "N61.18 W130.50".
# default: "unspecified"
# latlong "N32.87 W117.22"
#
# The URL for more information on the Cluster. Intended to give purpose,
# owner, administration, and account details for this cluster.
# default: "unspecified"
# url "http://www.mycluster.edu/"
#
# The location of this host in the cluster. Given as a 3D coordinate:
# "Rack,Rank,Plane" that corresponds to a Euclidean coordinate "x,y,z".
# default: "unspecified"
# location "0,0,0"
#
# The multicast channel for gmond to send/receive data on
# default: 239.2.11.71
# mcast_channel 239.2.11.71
#
# The multicast port for gmond to send/receive data on
# default: 8649
# mcast_port    8649
#
# The multicast interface for gmond to send/receive data on
# default: the kernel decides based on routing configuration
# mcast_if  eth1
#
# The multicast Time-To-Live (TTL) for outgoing messages
# default: 1
# mcast_ttl  1
#
# The number of threads listening to multicast traffic
# default: 2
# mcast_threads 2
#
# Which port should gmond listen for XML requests on
# default: 8649
# xml_port     8649
#
# The number of threads answering XML requests
# default: 2
# xml_threads   2
#
# Hosts ASIDE from "127.0.0.1"/localhost and those multicasting
# on the same multicast channel which you will share your XML
# data with.  Multiple hosts are allowed on multiple lines.
# Can be specified with either hostnames or IP addresses.
# default: none
# trusted_hosts 1.1.1.1 1.1.1.2 1.1.1.3 \
# 2.3.2.3 3.4.3.4 5.6.5.6
#
# The number of nodes in your cluster.  This value is used in the
# creation of the cluster hash.
# default: 1024
# num_nodes  1024
#
# The number of custom metrics this gmond will be storing.  This
# value is used in the creation of the host custom_metrics hash.
# default: 16
# num_custom_metrics 16
#
# Run gmond in "mute" mode.  Gmond will only listen to the multicast
# channel but will not send any data on the channel.
# default: off
# mute on
#
# Run gmond in "deaf" mode.  Gmond will only send data on the multicast
# channel but will not listen/store any data from the channel.
# default: off
# deaf on
#
# Run gmond in "debug" mode.  Gmond will not background.  Debug messages
# are sent to stdout.  Value from 0-100.  The higher the number the more
# detailed debugging information will be sent.
# default: 0
# debug_level 10
#
# If you don't want gmond to setuid, set this to "on"
# default: off
# no_setuid  on
#
# Which user should gmond run as?
# default: nobody
# setuid     nobody
#
# If you do not want this host to appear in the gexec host list, set
# this value to "on"
# default: off
# no_gexec   on
#
# If you want any host which connects to the gmond XML to receive
# data, then set this value to "on"
# default: off
# all_trusted on

# gmond --conf=/etc/gmond_test.conf

4. Commandline Tools

Last Updated: $Date: 2002/10/16 18:35:13 $

prompt> gmetric --help
gmetric 2.5.0

Purpose:
  The Ganglia Metric Client (gmetric) announces a metric
  value to all Ganglia Monitoring Daemons (gmonds) that are listening
  on the cluster multicast channel.

Usage: ganglia-monitor-core [OPTIONS]...
   -h         --help                  Print help and exit
   -V         --version               Print version and exit
   -nSTRING   --name=STRING           Name of the metric
   -vSTRING   --value=STRING          Value of the metric
   -tSTRING   --type=STRING           Either string|int8|uint8|int16|uint16|int32|uint32|float|double
   -uSTRING   --units=STRING          Unit of measure for the value e.g. Kilobytes, Celcius
   -sSTRING   --slope=STRING          Either zero|positive|negative|both (default='both')
   -xINT      --tmax=INT              The maximum time in seconds between gmetric calls (default=60)
   -cSTRING   --mcast_channel=STRING  Multicast channel to send/receive on (default='239.2.11.71')
   -pINT      --mcast_port=INT        Multicast port to send/receive on (default=8649)
   -iSTRING   --mcast_if=STRING       Network interface to multicast on e.g. 'eth1' (default='kernel decides')
   -lINT      --mcast_ttl=INT         Multicast Time-To-Live (TTL) (default=1)

prompt> cputemp
63

prompt> gmetric --name temperature --value `cputemp` --type int16 \
--units Celcius

prompt> gstat --help
gstat 2.5.0

Purpose:
  The Ganglia Status Client (gstat) connects with a
  Ganglia Monitoring Daemon (gmond) and output a load-balanced list
  of cluster hosts

Usage: gstat [OPTIONS]...
   -h         --help             Print help and exit
   -V         --version          Print version and exit
   -a         --all              List all hosts.  Not just hosts running gexec (default=off)
   -d         --dead             Print only the hosts which are dead (default=off)
   -m         --mpifile          Print a load-balanced mpifile (default=off)
   -1         --single_line      Print host and information all on one line (default=off)
   -l         --list             Print ONLY the host list (default=off)
   -iSTRING   --gmond_ip=STRING  Specify the ip address of the gmond to query (default='127.0.0.1')
   -pINT      --gmond_port=INT   Specify the gmond port to query (default=8649)

prompt> gstat
CLUSTER INFORMATION
       Name: unspecified
      Hosts: 97
Gexec Hosts: 73
 Dead Hosts: 0
  Localtime: Mon Apr 22 16:58:43 2002

CLUSTER HOSTS
Hostname                     LOAD                       CPU              Gexec
 CPUs (Procs/Total) [     1,     5, 15min] [  User,  Nice, System, Idle]

mm92.millennium.berkeley.edu
    4 (    1/   97) [  1.10,  1.19,  0.99] [   5.9,   0.0,   0.5, 100.0] ON
mm98.Millennium.Berkeley.EDU
    4 (    0/   80) [  1.16,  1.67,  1.25] [   4.1,   0.0,   0.2,  98.5] ON
mm91.Millennium.Berkeley.EDU
    4 (    1/   87) [  1.67,  1.78,  1.69] [  25.0,   0.0,   0.7,  74.9] ON
mm75.millennium.berkeley.edu
    4 (    3/  103) [  1.85,  2.54,  1.83] [  72.6,   0.0,   0.2,  50.3] ON
mm67.millennium.Berkeley.EDU
    4 (    4/  112) [  1.89,  2.08,  1.38] [  81.4,   0.0,   0.1,  38.5] ON
mm87.millennium.berkeley.edu
    4 (    4/  112) [  1.95,  1.67,  1.27] [   3.2,   0.0,   0.4,  96.4] ON
mm83.millennium.Berkeley.EDU
    4 (    1/  120) [  2.00,  2.59,  2.24] [  25.0,   0.0,   0.0,  75.0] ON
mm10.millennium.Berkeley.EDU
    2 (    0/   77) [  0.00,  0.06,  0.07] [   0.2,   0.0,   0.0,  99.9] ON
...

prompt> gstat --mpifile
mm56.Millennium.Berkeley.EDU:4
mm44.Millennium.Berkeley.EDU:4
mm31.Millennium.Berkeley.EDU:2
mm43.Millennium.Berkeley.EDU:4
mm15.Millennium.Berkeley.EDU:2
...

5. Getting Help

If you need help, a good starting point is the latest ganglia documentation. If the manual doesn't answer your question, then the next place to look is the mailing list archives. It is very possible that another ganglia user has had the same problem as you. If you still haven't found an answer, then send mail to the ganglia-general@lists.sourceforge.net.

To report bugs, submit patches or ask questions of the ganglia developers, send an email to ganglia-developers@lists.sourceforge.net.

6. ChangeLog

2002-08-23 Federico Sacerdoti
   * Increased maximum gmetric value size to 1472 chars from 255. 
     (1472 = 1500 Eth MTU - IP and UDP header bytes) Better memory
     efficiency by storing only necessary bytes in hash.

2002-08-21 Matt Massie
   * Added gmetad to the monitoring core distribution

2002-08-16 Federico Sacerdoti
   * Added disk_free, disk_total, and part_max_used disk metrics for Linux.
     They do not double count automounted directories as 'df' sometimes
     does. 
   * Added LOCATION attribute to HOST tag for physical view of cluster.
   * Enabled the use of hostnames in gmond's trusted hosts list.

2002-08-14 Matt Massie
   * Updated gmetric to send the SLOPE and TMAX attributes to gmond
   * Making the XML port more bullet-proof when a remote client closes its
     connection prematurely
   * Changed the XML DTD to handle cases where a CLUSTER has no HOST or
     a HOST has no METRICs to report

2002-08-12 Matt Massie
   * Added TN (Time Now = number of seconds since last msg received) and
     TMAX (Time Max = maximum distance between multicast messages) to the
     HOST and METRIC XML elements.  Also added the SLOPE attribute to 
     METRIC which can be set to "zero", "positive", "negative" or "both" in 
     order to make the Round-Robin database structure more efficient

2002-08-08 Preston Smith
   * Fixed AIX/libdnet build problems. (Turned out to be header file 
     order in gmond.c and gmetric.c)

2002-08-06 Federico Sacerdoti
   * Added a "timestamp" type to gmond and gmetric. Also changed some
     unit names. Added a OWNER and LATLONG attribute to the CLUSTER tag,
     both set in the conf file.

2002-07-24 Matt Massie
   * Stitched in the libdnet into the ganglia distribution to handle the
     low-level networking necessary for dynamic routing and interface lookup. 

2002-07-05 Preston Smith
   * Added a patch from Davide Tachella to get CPU MHz on AIX, along
     with appropriate autoconf magic to detect it, and tell how to get
     the libraries that report it if they're not found.

2002-06-26 Matt Massie
   * Fixed the heap corruption and thread madness on Solaris boxes.

2002-06-24 Preston Smith
   * Merged linux-alpha platform specific code into linux.c
     (cpu # and cpu MHz are the only different metrics)
     Fixed configure.in to reflect this.

2002-06-21 Matt Massie
   * Completely rewrote the underlying network library in order to allow
     for future growth (e.g. automatic discovery and all network interfaces,
     sending/receiving on multiple network interfaces, multicast forwarding)
     and to fix known bugs

2002-06-20 Preston Smith
   * Tweaked cpu_speed_func to only use the existing code for
     i386, ia64, and hp-parisc.
     Added code to get cpu clock speed for powerpc
     Sparc, s390, and arm don't report their cpu clock rate.
   * Added lines to machine_type_func to report arch name for
     sparc, alpha, powerpc, m68k, mips, arm, hppa, s390
     Couldn't log into an m68k or mips box to check them out, so they may
     or may not successfully run gmond.

2002-06-20 Preston Smith
   * Added patches to aix.c to support AIX 5.
   * Added FreeBSD to the list that has SUPPORT_GEXEC=0 by default 
     (configure.in)

2002-06-17 Steven Wagner
   * Updated autoconf to set SUPPORT_GEXEC flag if building on a supported
     platform (use --enable-gexec).  gmond should report full IRIX metrics
     now, and run "out of the box."

2002-05-31 Matt Massie
   * Fixed a bug reading the XML output where some hosts had timestamps newer than
     the cluster timestamp and were mistakenly marked as down

2002-05-13 Matt Massie
   * Changed pre_process_node() gethostbyaddr() error reporting to be more sane 
     for environments which nameservice problems
   * Updated listen.c to prevent out-of-range multicasted key values from crashing
     gmond in debug mode 

2002-05-06 Matt Massie
   * Added an explicit heartbeat metric to gmond along with a GMOND_STARTED
     HOST attribute which removes the need for a 90 second delay in gmond restarts
   * Updated linux.c to correctly support the ia64 architecture

2002-05-01 Matt Massie
   * Gmond now has a configuration file for its configuration instead of
     passing commandline arguments

2002-04-26 Asaph Zemach
   * Completely rewrote ./gmond/machines/linux.c to be much more efficient by 
     removing the need for 2 threads and pthread mutexes.

2002-04-26 Matt Massie
   * Modified net.c setsockopt() functions to be more portable

2002-04-19 Matt Massie
   * Fixed a bug in the .spec file where upgrades failed if gmond was not up
   * Updated autoconf to correctly configure ia64 machines

2002-04-18 Matt Massie
   * Enabled setsockopt SO_REUSEADDR for the mcast_join socket in order allow
     multiple instances of gmond to listen to the same multicast channel

2002-04-17 Matt Massie
   * Added the --list and --single_line option to gstat for output flexibility

2002-04-17 Alan Hagge
   * Added preliminary support for IRIX

2002-04-16 Matt Massie
   * Add the "gexec" metric in preparation for the release of gexec
   * Updated the gexec_cluster() function in libganglia to pull in more data
     from the XML and handle unresolved names and names without domains correctly
   * Updated gstat to print the updated information from gexec_cluster()
   * Added a --no_gexec flag to gmond for hosts that are not part of the
     computation cluster (file servers, frontends, etc)
   * Added a --all_trusted flag where gmond sees ALL hosts as trusted as
     suggested by Martin Knoblauch
   * Removed goto statement in pre_process_node() function to avoid rare looping bug
     thanks to feedback from Mike Snitzer

2002-04-12 Chris Elmquist
   * Updated gmetric.c to check if the name of the metric is empty to prevent
     the problem of blank metrics showing up in gmonds

2002-04-08 Matt Massie
   * Fixed a bug in the barrier code used in gmond
   * Fixed mcast_connect() on non-Linux systems related to the IP_MULTICAST_TTL option
   * Added the "--no_setuid" and "--setuid" flags to provide more euid flexibility

2002-04-05 Preston Smith
   * Added support for AIX machines

2002-04-04 Dave Wallace
   * Fixed a bug in gmond which caused big-endian architectures to incorrectly
     store multicast data and therefore misreport on the XML port.
   * Added a key value check of the XDR multicast data in ./gmond/listen.c 

2002-04-04 Matt Massie
   * Updated ./lib/gexec_func.c gexec_cluster() function to handle hosts with
     no domainname correctly (as pointed out by David Wallace
   * Updated the program commandline options to obey the GNU Coding Standard
   * Added a --debug_level parameter to gmond to allow debugging the daemon
     without recompiling the daemon 
   * Updated the --trusted_host option to allow multiple instances of the option
   * Added a --mcast_ttl gmond option to allow you to modify the 
     Time-To-Live (TTL) of the outgoing multicast messages

2002-04-02 Neil Spring
   * Updated the gmond Makefile to cleanup the machine.c link on "make clean"

2002-04-02 Doc Schneider
   * Update the way the number of CPUs are collected in order to workaround a 
     bug on AMD-based systems.

2002-03-28 Matt Massie
   * Change distribution tree to fit the ganglia model.  All gmond clients are
     in the ./gmond directory and all gregisterd clients in ./gregisterd

2002-03-26 Matt Massie
   * Removed the use of streams (via fdopen) on the XML socket and replaced
     it with write()s on the socket descriptor to work around Linux bug under
     high-stress conditions

2002-03-25 Matt Massie
   * Added thread barriers to gmond initialization to ensure listening threads 
     exist before the threads which multicast are created

2002-03-24 Matt Massie
   * Use XML_ParseBuffer() in gexec_cluster() in order to avoid
     double copying of buffers (XML input from gmond). Faster.

2002-03-22 Matt Massie
   * Updated gexec_cluster_free() to ensure no memory leaks even when
     cluster.num_nodes and cluster.num_dead_nodes equals zero
   * Used setvbuf to ensure the gmond and libganglia are using line buffering

2002-03-21 Matt Massie
   * Updated net.h to include netinet/in.h
   * Changed sockaddr_in_new function in net.c to plug a potential memory leak
   * Added XML_ParserFree() call to gexec_cluster() lib call to plug up 
     memory leak
   * Modified ./gmond/server.c to prevent crashes under heavy stress conditions
   * Added fclose() calls to gexec_cluster() to plug a memory leak
   * Added gexec_cluster_free() call to gstat for good measure

2002-03-15 Preston Smith
   * Patched a wrong sysctl to get free memory for freebsd machines

2002-03-15 Matt Massie
   * Added mute and deaf mode for gmond
   * Created a new ganglia-monitor-core-lib distribution for the
     libganglia library.
   * Moved all the documentation to DocBook, added much more information
     and updated/removed what was there.  Output docs to ./docs directory
     of the distribution in both HTML and PDF form.  Also installed
     Doxygen to document libganglia.
   * Add the Ganglia Status Tool (gstat) which allows you to check the status
     of your cluster from the commandline.  Hosts are sorted with least-loaded
     nodes at the top of the list.

2002-03-12 Matt Massie
   * Removed the need for the POSIX mutex in pre_process_node() allowing for
     faster processing of incoming multicast data

2002-03-11 Matt Massie 
   * Changed gmond to not count itself as a running process when reporting
     the number of running processes.

2002-03-06 Doc Schneider
   * Changed the way ganglia builds RPMs to support non-root builds as
     suggested by an anonymous SourceForge user

2002-02-28 Matt Massie 
   * Added a new CLUSTER element to the XML with two attributes:
     NAME and LOCALTIME.  Necessary for monitoring clusters in 
     many different timezones.

2002-02-27 Matt Massie 
   * Fixed the getopt_long() call in gmond.c thanks to feedback from 
     Meik Hellmund.  The getopt_long() parameters didn't match the 
     switch() statement breaking the "trusted_host" option.
   * Fixed a bug in gmond where connections from untrusted hosts caused
     segfaults.  Error caused by passing datum_free() a NULL pointer in
     server_thread() of ./gmond/server.c.
   * Changed the way transient nameservice errors are handled by 
     pre_process_node() in ./gmond/listen.c.  Previously, transient errors
     were retried but now they are treated as errors (although gmond 
     will continue trying to resolve the host when it gets a new 
     multicast packet from it)
   * Updated the ganglia.spec file to merge gmond and gmetric into a single
     RPM, fixed some small bugs, and updated the RPM information.
   * Changed the gmetric options to also support long options and updated
     the help output (from -h, --help) to be much more descriptive

2002-02-27 Preston Smith
   * Updated the FreeBSD monitoring code to include all metrics
     which are monitored under Linux except number of running processes,
     absolute cpu idle time, and shared memory.  SMP users may find that
     freebsd's cp_time sysctls is not completely accurate under FreeBSD stable
     meaning CPU%s might be inaccurate.  However, it works under FreeBSD-CURRENT.

2002-02-22 Matt Massie 
   * Added the getopt source to the ganglia library for system that don't 
     have the getopt API available (Solaris, FreeBSD, etc)

2002-02-21 Matt Massie 
   * Changed the "safe_host" option to "trusted_host" to make it clearer
   * Added the "num_nodes" and "num_custom_metrics" commandline options

2002-02-20 Matt Massie 
   * Completely rewrote the underlying hash library because the original
     hash functions where over-engineered and had some memory leaks.
     New hash functions are superlight and fast.  Built test program and
     profiled/traced all memory functions using mpatrol.  No leaks.
   * Updated code to catch when transient nameservice errors occur and retry.
     Correctly handle hosts the don't resolve instead of treating as an error.
   * Added a patch submitted by Joshua J England for gmond to correctly report
     the number of CPUs and their speed on alpha architectures 
   * Added a patch submitted by Eirikur Hallgrimsson and written by
     Yaroslav Klyukin for gmetric which allows users to chose which network 
     interface gmetric multicasts metric data

2002-02-12 Matt Massie 
   * Reduced the number of total threads by one by removing the 
     for(;;)pause() spin and having the main thread do server work

2002-02-07 Matt Massie 
   * created the function my_inet_ntop() function in libganglia to deal with
     the limitations of inet_ntoa in a multi-threaded environment
   * changed the self-organzing behavior of gmond to recognize when a transient
     error occured on a remote gmond process
   * added verbose error checking of gethostbyaddr() in listen.c

2002-02-04 Matt Massie 
   * fixed a bug in the cpu_num_func() which reported AMD systems as having
     twice as many CPUs as they really did

2002-02-01 Matt Massie 

   * increased the speed of the host security check for the XML port
   * added commandline options for almost all compile-time opts for gmond
   * added a --safe_host option to allow a host outside of the multicast
     channel to connect
   * now gmond strips all quotes (") from gmetric data to keep XML well-formed 
   * improved the self-organizing behavior of the gmonds

2002-01-18 Matt Massie 
  
   * changed the name of the distribution from ganglia to 
     ganglia-monitor-core to be more descriptive
   * modified ./gmond/server.c in order to keep gmond from crashing 
     when a client closes the connection prematurely
   * added patch to ./gmond/gmond.c from Neil Spring and Brian Youngstrom to 
     add command line option and properly setuid to nobody 
   * added gmond.init.SuSE to the distribution.  Contributed by Oliver Mossinger.

7. License

Ganglia is released under the following BSD license.

Copyright (c) 2001, 2002 by The Regents of the University of California.  
All rights reserved.

Permission to use, copy, modify, and distribute this software and its
documentation for any purpose, without fee, and without written agreement is
hereby granted, provided that the above copyright notice and the following
two paragraphs appear in all copies of this software.

IN NO EVENT SHALL THE UNIVERSITY OF CALIFORNIA BE LIABLE TO ANY PARTY FOR
DIRECT, INDIRECT, SPECIAL, INCIDENTAL, OR CONSEQUENTIAL DAMAGES ARISING OUT
OF THE USE OF THIS SOFTWARE AND ITS DOCUMENTATION, EVEN IF THE UNIVERSITY OF
CALIFORNIA HAS BEEN ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

THE UNIVERSITY OF CALIFORNIA SPECIFICALLY DISCLAIMS ANY WARRANTIES,
INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY
AND FITNESS FOR A PARTICULAR PURPOSE.  THE SOFTWARE PROVIDED HEREUNDER IS
ON AN "AS IS" BASIS, AND THE UNIVERSITY OF CALIFORNIA HAS NO OBLIGATION TO
PROVIDE MAINTENANCE, SUPPORT, UPDATES, ENHANCEMENTS, OR MODIFICATIONS. 

8. Notes

State of the IRIX port:

*  CPU percentage stuff hasn't improved despite my efforts.  I fear there
   may be a flaw in the way I'm summing counters for all the CPUs.
*  Auto-detection of network interfaces apparently segfaults.
*  Memory and load reporting appear to be running properly.
*  CPU speed is not being reported properly on multi-proc machines.
*  Total/running processes are not reported.
*  gmetad untested.
*  Monitoring core apparently stable in foreground, background being tested
(had a segfault earlier).

State of the Tru64 port:

*  CPU percentage stuff here works perfectly.
*  Memory and swap usage stats are suspected to be inaccurate.
*  Total/running processes are not reported.
*  gmetad untested.
*  Monitoring core apparently stable in foreground and background.

State of the Solaris port:
*  CPU percentages are slightly off, but correct enough for trending
   purposes.
*  Load, ncpus, CPU speed, breads/writes, lreads/writes, phreads/writes,
   and rcache/wcache are all accurate.
*  Memory/swap statistics are suspiciously flat, but local stats bear
   this out (and they *are* being updated) so I haven't investigated
   further.
*  Total processes are counted, but not running ones.
*  gmetad appears stable

Anyway, all three ports I've been messing with are usable and fairly
stable.  Although there are areas for improvement I think we really can't
keep hogging all this good stuff - what I'm looking at is ready for
release.

 Debian packages for Debian 3.0 (woody) are available at
  http://www.physics.purdue.edu/~psmith/ganglia
 (i386, sparc, and powerpc are there presently, more architectures will
  appear when I get them built.)
 Packages for "unstable" (sid) will be available in the main Debian
  archive soon.

 Also, a CVS note: I checked in the debian/ directory used to create
 debian packages.

i need to add a section in the documentation talking about this since it 
seems to be a common question.

when you use...

mcast_if eth1

.. in /etc/gmond.conf that tells gmond to send its data out the "eth1"
network interface but that doesn't necessarily mean that the source
address of the packets will match the "eth1" interface.  to make sure that
data sent out eth1 has the correct source address run the following...

% route add -host 239.2.11.71 dev eth1

... before starting gmond.  that should do the trick for you.

-matt

> I have seen some post related to some issues
> with gmond + multicast running on a dual nic
> frontend.
> 
> Currently I am experiencing a weird behavior
> 
> I have the following setup:
> 
>   -----------------------
>   | web server + gmetad |
>   -----------------------
>              |
>              |
>              |
>     ----------------------
>     |   eth0 A.B.C.112   |
>     |                    |
>     |  Frontend + gmond  |
>     |                    |
>     | eth1 192.168.100.1 |
>     ----------------------
>              |
>              |
> 
>        26 nodes each
>           gmond
> 
> In the frontend /etc/gmond.conf I have the
> following statement: mcast_if  eth1
> 
> The 26 nodes are correctly reported. 
> 
> However the Frontend is never reported.
> 
> I am running iptables on the Frontend, and I am seing
> things like:
> 
> INPUT packet died: IN=eth1 OUT= MAC= SRC=A.B.C.112 DST=239.2.11.71 
> LEN=36 TOS=0x00 PREC=0x00 TTL=1 ID=53740 DF PROTO=UDP SPT=41608 DPT=8649
> LEN=16 
> 
> I would have expected the source to be 192.168.100.1 with mcast_if eth1
> 
> Any idea ?

Perhaps information regarding gmond on networks set up through cisco
catalyst switches should be mentioned in the ganglia documentation. I think
by default multicast traffic on the catalyst will flood all devices unless
configured properly. Here is a relavent snipet from a message forum, with a
link to cisco document.

--
If what you are trying to do, is minimizing the impact on your network due
to a multicast application, this link may describe what you want to do:
http://www.cisco.com/warp/public/473/38.html

We set up our switches according to this after a consultant came in and
installed an application multicasting several hundred packets per second.
This made the network functional again.