Monitoring server room temperature

Question

I have a small server room with its own AC unit. Recently, the AC died, and the temperature increased from 70 F to > 90 F. We rarely go in this room, so I was lucky that someone happened to notice that the fans were running a lot louder than normal as they walked past the door. It looks like I need a way to be notified when the temperature in that room gets too hot.

What tools are you using to monitor the temperature in your server room? How does this tool notify you of a problem (email, SNMP, etc).

Note: I've read this question on server temperature, but I'm interested in the whole room, not just the inside of a server case.

Edit:

Thanks for all the great responses so far! Many of these products measure much more than just temperature. What else should I be looking at and why?

Solution 1

Take a look a ITwatchdogs. Their weather goose line looks very nice. They monitor temp, light, sound, humidity, etc.

Here's a list of vendors (not APC) that have other products.

What I use in my data center appears to be a discontinued model. What ever you end up using, make sure that:

• They use SNMP, not some special protocol.
• Have remote temperature sensors. (I don't know if the netbotz do ...)

Currently the temperature monitors are hung from the middle of the drop ceiling.

Using a simple network monitoring application (zenoss, munin, nagios, etc) just monitor for threshold violations and create alerts for your pager/email.

You should also look into buying a handheld thermometer and walk around to every part of your datacenter/comm closet (including corners) to find any hotspots.

Solution 2

A couple of specific answers and a suggestion ...

1- There are several makers (linked in other answers) of standalone temp and humidity monitoring devices. These are pretty simple .. typically you mount them where you want to monitor, plug them into the network, and configure them to email alerts. They work well.

2- Most servers can raise SNMP alerts when the temperature gets out of spec, and you can use something like WhatsUp from Ipswitch to accept and do something about those alerts.

You should make a point to visit the server room once or twice a day, perhaps on arrival and after lunch, to check on the temp and humidity as well as examine the equipment for lights or status codes that are not normal, unusual noises, etc. If you can't do it then get someone else to do so.

Solution 3

If you want a cheap solution, you can buy USB thermometers for about 10USD. I've got a TEMPer device. I've not got around to using it yet, mostly because it's not entirely supported under Linux. Tollef Fog Heen wrote some support, but I haven't checked to see if it's in the kernel yet. It does ship with Windows drivers, so it just a matter of hooking it up to your monitoring. :)

Solution 4

+1 to Joseph Kern's answer. We use one of ITWatchdog's systems,but I've used other systems as well, and the general idea is great.

ITWatchdogs, and a lot of other vendors, make use of the Dallas 1-wire bus to drive their sensors. The actual sensors are available from Farnell etc for very low prices, and they basically attach to a piece of cat3 cable. You can combine an ITWatchdogs (or similar) system with a lot of hand-assembled 1-wire sensors for a relatively cheap solution.

There are also standard RS232 interfaces to the 1-wire bus, (such as this one) with varying levels of complexity, which mean you can directly attach monitoring gear to servers, in case you don't want to run additional gear (remote POP with only one machine, for example)

Solution 5

We use the MiniGoos products from http://www.itwatchdogs.com/. I will admit I wasn't involved in the evaluation process for these, but I really like the product. They allow multiple temperature sensors and they have some nifty add ons like air flow monitoring (which we aren't currently using) and things like door open/close detection (which we are). They have an HTTP interface which could be scraped if you wanted, but also support SNMP traps and polls. We plugged it into nagios and it worked like a champ.

I also second and extend a comment josephkern made in his answer. In addition to walking around with a thermometer to get a feel for the temperature pattern in the room, buy some cheap non-connected thermometers and hang them from the ceiling. They're great backup for reporting thermometers, and you'd be surprised how often a problem will be spotted by a human glancing at the therm as they walk in and out of the room before the warning range of the monitoring thermometers is crossed. The earlier you diagnose an AC/HVAC problem, the longer you will have to fix it before you start damaging components, having to shut down machines, or having machines shut down on their own to protect themselves.

View more solutions

Comments

[Oskar Duveborn]

Someone to cover your back for you when not available, temperature sensor wired to the building/office security alarm for instance...

[wfaulk]

Only 90°F? Mine got up to 140°F yesterday. :(

[Matt Rogish]

+1 this is what we use. Works great and we know that someone will CALL the phone list when it goes off (no worrying if my phone is turned off or silent, etc.)

[Andrioid]

+1 for Sensatronics. Works perfectly and then you just use your server monitoring solution for checking out the SNMP device (and graph it, if you please).

[Wegged]

any links to any documentation ?

[Jon Angliss]

I actually did something similar a while back... jon.netdork.net/2009/07/07/tinkering-with-temperature and jon.netdork.net/2009/08/01/tinkering-with-temperature-part-d‌​eux. Then the DS18B20 are attached to some CAT5 cables, and can be run to wherever you want in the server room. Box is connected to the back of a server (this one is running Nagios), which pings the probes, and grabs the temp.

[avatar1234]

Agreed, this is what we use as well. Since the alarm is battery protected and through separate communications than the IT infrastructure, it covers a few more cases than a system that depends on the circuits and devices that it is protecting. It's also a very inexpensive solution.

[Olli]

In general, you are probably doing something wrong if you need alarm system often.

[Day]

The perl script given in this blog post works with my TEMPer thermometer on Ubuntu 10.04

[jqa]

1-wire is a great budget method and can monitor lots of things. We were about to buy some 1-wire kits and try to figure out the software when we found a Netbotz 310 for $40 on ebay that did it all plus had good software and management tools.