25 May 2013 elwell   » (Journeyer)

Low-Level hardware hacking

In our 3 shiny new computer rooms (supercompute, IO, Tape - Each with slightly different environmental and cooling requirements), the architects have fitted a single temperature sensor on the back wall. Out of the airflow. This isn't giving us a terribly accurate picture of the supply air going into each rack, and so an alternative DIY method is underway.

A Raspberry Pi in each room (small, low enough power that it can be fed from PoE via a splitter) acts as the local 'head' and grabs the data to send it to a central message broker. From there, we can choose what to do and how to display it.

The sensor hardware consists of: A single DHT22 attached near to the Pi - As useful as these are on an arduino, on a R-Pi they are a bit harder due to the need for timing. The Adafruit interface for these grabs directly from memory address and seems well, a bit hacky. As suggested on IRC, a proper kernel module may be helpful. hint for someone else to complete :-)

The main sensor array is a set of DS18B20 sensors from fleabay (cheap if you're ordering 50 at a time) attached to the sheepwalk i2c interface 'RPI2'. These are then strung around to suit depending on location (ie, for our watercooled racks I'm going to affix the sensors directly to the flow/return lines on the watercooled racks to give us the delta-T across each set of door radiators (The BMS only gives flow/ret temp overall) which when tied to the internal temp sensors in the back of the racks should give us an idea of how well we're shifting the heat away from the machines. All this hardware comes in at less than 100 AUD per room - and some soldering time to make up the sensor chains to suit. To get an idea of the stratification (and therefore how do we position floor tiles relative to the overhead extraction vents) I'm going to put 15-20 sensors on a piece of plastic waste pipe to get a nice vertical profile - cue trip to bunnings (local DIY store) later this weekend.

Software

Hat-tip to the excellent MQTT people - nice n simple to use a python publisher script to send to a mosquitto broker on the lan. From there we can republish to Xively (useless as it doesn't respect timezones in the plotting so I can only see data I sent 8h ago...), Google spreadsheet (in a refactor of the original adafruit idea, and currently to a simple python curses console
Once we go into production and kit is running properly I plan to use the broker for all the machine status info and we'll dispatch accordingly for alarming.

More info and code will appear at https://github.com/ivecops/EnvMon

Syndicated 2013-05-25 07:06:00 (Updated 2013-05-25 07:06:57) from Andrew Elwell

Latest blog entries     Older blog entries

New Advogato Features

New HTML Parser: The long-awaited libxml2 based HTML parser code is live. It needs further work but already handles most markup better than the original parser.

Keep up with the latest Advogato features by reading the Advogato status blog.

If you're a C programmer with some spare time, take a look at the mod_virgule project page and help us with one of the tasks on the ToDo list!