Real time monitoring of MS Windows servers and their services

Question

We have a bunch of large HDTVs in our monitoring office keeping an eye on all of our production equipment.

We are monitoring:

Cisco routers

• HP switches
• HP proliant servers
• Windows 2003
• IIS
• SQL server

At the moment we use

• Nagios for uptime/availability and alert sending
• Cacti for bandwidth usage
• Perfmon running on Vista for server performance
• A combination of other tools and our own custom code to monitor our actual application performance.

All of this is fine apart from the Perfmon part - it gives us what we want - i.e. real time charts on the screen, logging certain performance counters, etc - the only problem is setting it up is a real chore. If the Vista PC running Perfmon is rebooted (normally due to Windows Update) then setting all the counters up again takes ages - literally an hour or two's worth of work for somebody in the office...

Anybody know of a way to either: 1. Script adding Perfmon counters 2. Another tool with graphical output and WMI/windows counter access.

Thanks

• Mike

Solution 1

I would seperate out the charting/display from the montioring/alerting piece. You haven't said how many object you have. For most shops I'd write a powershell script to get the counters I want and pipe the output to powergadgets and for small shops use polymon to monitor the windows stuff and groundwork to monitor the rest. In a larger environment (30+ servers) I'd take a look at System Center Operations Manager for monitoring and alerting. You have to realize howver that SCOM is NOT just monitoring. It's designed to correct problems and ensure business service, unlike most monitoring software that only cares about server counters. Other options would be tivoli or openview.

Solution 2

I highly recommend Zabbix to system administrators looking for monitoring and graphing solutions.

Zabbix has clients (agents) for Windows and Linux systems. They can also read SNMP packets to graph and monitor stats from just about anything.

I've been using it for 3 years and it has been (relatively) very easy to manage and keep updated.

They even have application monitoring if you are willing to script the actions for it.

_{(source: zabbix.com)}

Let me know if you have any specific questions about usage.

Solution 3

OpenNMS 1.7.x has WMI support. OpenNMS

Otherwise I would go the Nagios/RRD route as joseph suggested.

Solution 4

Perfmon can be scripted, and the settings can be saved in Vista. Perfmon can even logged to a MSSQL database. Here's how to do it.

1. Setup a database server
2. Create an ODBC connection on the system which you are running perfmon
3. Create a new perfmon counter and direct it towards the ODBC connection
4. Startup perfmon, and watch it log

PROTIP: You can setup your single vista machine to also remotely log perfmon data from other systems.

And here's how you can generate a report or two from the data you've collected.

Solution 5

if not realtime charts [ that is with few minutes sampling time ] are ok for you - take a look at Munin Node for Windows. alternatively you can probably fetch plenty of statistics from windows via snmp but i've tried it long time ago and abandoned idea quite quickly. munin-node was just enough.

View more solutions

Comments

[Mike McClelland]

Whilst I haven't finished this exercise - I am most likely to use PowerGadgets... Looks good, works the way I want, etc. Will post back any solutions to this page. Thank you very much to everybody who's responded - what an awesome community! Mike

[Mike McClelland]

Hi RennieJ. Yes, you can - BUT if the machines you are monitoring are not local to the domain of the monitor PC (I think is the reason) - then that technique doesn't work - the .msc opens but doesn't connect. It used to on Perfmon XP, but Perfmon on Vista is different - and apart from this 'feature' is much better.

[Mike McClelland]

Hi pQd - looks great - I'd struggle with getting permission to install the client app on our production kit! Seems silly to use something 3rd party when 'it's all there' in the Windows Framework...

[John Rennie]

Hmmm, possibly an authentication problem? If you do a net use \\target\ipc$ before opening the .msc does that make any difference?

[Mike McClelland]

Yeh, we had a batch file that connected to all the shares and get's a handle BEFORE opening PerfMon. Under XP this worked, but not the Vista version... We'd revert to XP, but the Vista version has other benefits - i.e. the .MSCs now remember exact layout, and perfmon now supports limits other than 100 - so you can get better granularity of perf when for example CPU is 0-25% - which it normally is...

[John Rennie]

>perfmon now supports limits other than 100 Hmmm, W2k3 perfmon supports this. Are you sure XP doesn't (I don't have an XP handy to check)? Just right click, go to the Graph tab and change the graph maximum.

[Joseph Kern]

You don't even need to set up the database server ... But in my experience dealing with spreadsheets on a monthly basis is a real pain. Why not use Nagios? I tried SCOM, hated it. Then switched to Zenoss (Nagios offshoot) and love it. Your Nagios rrd graphs will transfer to Zenoss and everything. zenoss.com There's a paid and a free version of zenoss available. Good luck!

[Toby]

you don't need to install anything, you can perfectly monitor a windows with snmp (in combination with munin) munin.projects.linpro.no/wiki/HowToMonitorWindows

[raja]

SNMP won't give you acces to any raw perfmon counters you might need. (eg you want to create a task manager type display), but you shouldn't need to actually install any software agents as long as WMI is accessable

[raja]

How does Zabbix read wmi counters? Searching the manual doesn't even list WMI. It also appears to not run on windows.

[Mike McClelland]

Hi Chris, yes agreed it can do that but the problem is getting the handles to the servers. We can open the C$ share, even open files on the remote FS, but PerfMon refuses to accept that the handle is there and wants you to make a new one - which takes about 90seconds per counter - and we have lots of servers and counters! Mike

[plaisthos]

Yea, I can see the time delay, but for me its about 30 secs. That's kind of a bummer.

[Dan Esparza]

Looks like it might actually be www.samurize.com

[Mike McClelland]

Hi Marco, that's an awesome answer - thanks very much. Only problem is that I don't really want to install anything on the servers that is neither our application of the OS (and any OS supporting stuff)... I'll check out the NSClient though and post back here if I have any comments/feedback. Thanks again for your time. Mike

[Maxwell]

Zabbix agents do, you can query the WMI in vbs through the agent.