Our sensor network is two separate runs off of a (http://www.aag.com.mx) TAI8595 hub.
One run goes around the ceiling and to the air handler sensors. We have sensors in each quadrant of the room, and a humidity/temp sensor in the center. We placed our sensors in the center of each quadrant of the room, in order to stay away from the low flow areas near the walls and from doors to the outside. Our sensors are DS18S20's and DS2438's (from AAG as well.) we ordered pre-made jumpers from http://www.connectworld.net/cgi-bin/iec/home.html We had to special order them as plenum rated.. and we ordered blue colored cables and hoods to make sure that no one mistook them for ethernet. (Don't forget that the AAG connectors are 6 pin, not 4 or 8 which are much more common.) We did some early tests of things like distance and "accuracy"... to confirm that we could go as far as we needed to. We put a bunch of them on a loop in an icechest, with a small electric fan inside.. The heat from the motor gave us a "changing" temperature to measure and the fan kept the air moving. We left the covers off of the sensors to get the fastest settling time. Even with this attempt at an isothermal environment, there was almost a 3 degree difference between the lowest and highest readings. (Good enough for our purposes.) We did punch some more holes in the covers of the sensors because there wasn't enough flow past the sensor itself. The time constant was too long otherwise. We then put sensors all over a rack to see if we could make sense of the readings.. and the answer was "not really".... There was some variability based on where we put the sensors.. but not enough to say "we are having a problem with the left most fan on the SunXXX...". Factors such as cable placment, how "full" the back area of the rack was.. and whether or not the back door was open, ventilated, solid had much more effect on the outcome. About the best overall look at what a rack was doing was achieved by mounting the sensor with a tie wrap to the grill on the underside of the fan in the top of the rack. Our racks are full of all sorts of things, from "salvaged" Sun desktops to dual processor AMD 64's in 1u chassias. The fans in the units do a lot of air mixing, making the change in a given reading hard to relate to what caused the change.. and other factors just swamped the "small signal" changes due to things like stalled fans. Most of what we see is temperature traces which march in parallel fashion across the graph.. with overall shifts that match what happens with the room temperature.. There will be an occasional "baseline shift" which is usually due to opening or closing a rack back door. The biggest factor in the change is "what's the room temp". We do see about a 2 degree change when the tape robot lights off for the nightly backup run in the ceiling sensor just above the robot. Our ceiling temps vary across the room by about 5.5 degrees from coolest to warmest.. And the delta T across the air handlers runs from a low of -2.5 degrees on a lightly loaded day to a high of 6.5 degrees under heavy load. Humidity goes from a low of 34 to a high of 62.. tracking nothing I've been able to spot.. except that it's highest on days of low delta T... which makes sense.. it's hard to remove moisture from a room that's already cold enough. We don't do reheat for humidity control. (We could.. but since we don't have the "problem" we don't want to pay high energy bills to solve the nonexistent problem. :-)) If I had lots and lots of racks to do, I'd probably run the sensor cable across the top of the racks, tie wrapped to the top grill of the fan.. you will get the fan motor heat added to the heat coming from the machines in the rack... but it's not all that significant.. and you can run short jumpers from one rack to the other, instead of going down to the floor and all the way back up again in the adjacent rack. Entering air temp matched the air handler exit temperature to such a great extent that we didn't bother to put a sensor at the bottom of the racks. Alarms are done by scanning the rrd database for high readings.. we look for three high readings before we page.. either three high in one scan (the whole room is getting hot).. or three high in three different scans.. "one rack is getting hot" before we fire off a page. It cut our false alarms to ZILCH.. we do have an an occasional "spike" in a reading for reasons we have not been able to correlate with anything in particular. Spikes went WAY down after we solved this problem: http://www.wes.net/construction_remediation_services-zinc_whisker_detail.htm (I'm not endorsing this company... just using their web site.. We used a local contractor.. and replaced the entire raised floor in the computer room. We did do surveys using a scanning electron microscope to tell when the cleanup was adequately done.. and it was a MESS.. We had MANY power supply failures while this mess was going on.. it's a real threat... with real results. At one point we were loosing a couple of power supplies a week.. Fortunately it was rare to loose both supplies on a redundant supply box.) Watch the airflow patterns of dissimilar machines.. some draw cool air from the front and exhaust hot air out the back.. some draw from the sides... you can end up with "closed loops" forming between adjacent machines that are hotter than you expect. and mixing with cooler air by the fan at the top can hide this result. Mixed racks may be good candidates for "more than one sensor". >From just one sensor per rack.. and other "environmental" sensors arrayed around the room we get pretty reasonable results, without spending a fortune to "over sense" a chaotic world for no gain. Steve ------------------------------------------------------- This SF.Net email is sponsored by: NEC IT Guy Games. How far can you shotput a projector? How fast can you ride your desk chair down the office luge track? If you want to score the big prize, get to know the little guy. Play to win an NEC 61" plasma display: http://www.necitguy.com/?r=20 _______________________________________________ Owfs-developers mailing list Owfs-developers@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/owfs-developers
