I have sysstats already but want a graphing.  ksar sounds appealing. I do not have physical access to this server so I need to access ksar through http on the server.  Does it support that?  Googling for the answer on how to do this but not finding what I want yet.

No, kSar is a desktop app and is IMO a PITA to use - I keep it in my pocket only when I need very fine granularity.

Most of the other ones mentioned have web front ends - Munin being the one I went with but I evaluated most of the other ones mentioned.

On 4/26/2010 11:38 AM, Matt wrote:

...

Take a look at the "sysstat" package which will collect data over time. There's a java-based desktop app called "kSar" that can use this data to generate graphs.

I have sysstats already but want a graphing. ksar sounds appealing. I do not have physical access to this server so I need to access ksar through http on the server. Does it support that? Googling for the answer on how to do this but not finding what I want yet.

Physical access shouldn't be necessary. Snmp based tools expect to work over the network and you can run local graphic programs remotely through an ssh tunnel. If you have to use only http, you could run cacti on the server itself, monitoring itself - and it is one of the easier systems to set up.

On Mon, Apr 26, 2010 at 12:13 PM, Brian Mathis brian.mathis@gmail.com wrote:

Take a look at the "sysstat" package which will collect data over time.  There's a java-based desktop app called "kSar" that can use this data to generate graphs.

I use kSar as well as Munin. Munin's only problem is that it has a hard coded 5 minute granularity - which is perfectly fine for 99% or more of what you need. I run this script on my systems which keeps sar logs with finer granularity that I can check into with kSar if I need to. But I almost never need to.

Run this every hour from cron to keep sar files for each hour

01 * * * * /root/sadc.sh

You may choose to only turn this on at times when you need that kind of granularity - below is pretty extreme with 10 second granularity. But even at that I do not find it impacts performance in any noticable fashion.

#!/bin/bash

# DATEFORMAT="%Y.%m.%d %H:%M:%S" HOSTNAME=`hostname | awk -F. '{print $1}'` DATEFORMAT="%Y.%m.%d %H:%M:%S" # SAINTERVAL x SACOUNT = 3600 SAINTERVAL=10 # seconds SACOUNT=360 # iterations TIMESTAMP=`date +"${DATEFORMAT}"` LOGFILEMAX=80 # max number of files to store ((LOGFILEMAXMINUTES=60*LOGFILEMAX)) LOGDIR="/var/log/sa/sadc" mkdir -p $LOGDIR if [ ! -d $LOGDIR ] then echo "ERROR: $LOGDIR does not exist" exit 1 fi LOGFILE=`echo "${LOGDIR}/sadc_${HOSTNAME}_${TIMESTAMP}" | sed "s/ /_/g" `

# are we 32 or 64 bit?

BASEARCH=`uname -i` case $BASEARCH in i386) SADC="/usr/lib/sa/sadc" ;; x86_64) SADC="/usr/lib64/sa/sadc" ;; *) echo "ERROR: unknown architecture [$BASEARCH]" exit 2 esac

$SADC -d -I -F $SAINTERVAL $SACOUNT $LOGFILE

# get the date from LOGFILEMAXMINUTES minutes ago and then touch # a file so it has that date. We'll then use this file # in the "find" command to find all files older than # it, so we can remove them

MAXFILESTAMP=`date -d "$LOGFILEMAXMINUTES minutes ago"` MAXFILENAME=syscheckmaxfile MAXFILE="${LOGDIR}/${MAXFILENAME}" touch -d "${MAXFILESTAMP}" $MAXFILE

pushd $LOGDIR RMFILES=`find . ! -name $MAXFILENAME -type f ! -newer $MAXFILE` for afile in $RMFILES do rm -f $afile done popd :

On 4/26/2010 11:15 AM, Marcelo M. Garcia wrote:

Matt wrote:

...

Is there a package I can get that will graph system resources such as CPU and disk I/O to an html file or something?

Matt

Hi

I recommend Ganglia. http://ganglia.sourceforge.net/

If you are using for a single system can be an overkill.

Likewise, if you have enough systems to make it worth setting up, OpenNMS is good and will also monitor your network equipment: http://www.opennms.org.