[In a message on Thu, 09 Jun 2005 16:31:22 PDT, ""Wright, Ryan P"" wrote:] >I could use a little help with the design of my 1-wire network. My >intention is to build a temperature sensing grid throughout a data >center using several hundred sensors. >
Several hundred sensors? Most data centers will run out of floor space before you can put that many racks in the room! We haven't seen much "value" in a sensor per machine.. the air is so mixed by the fans in the box that you really can't predict much about what is going on with a given machine in the rack. You don't want to put sensors inside the machines, because you will block, a possibly "significant", amount of airflow with the sensor. Our best luck has been a sensor per rack and a "grid" of sensors on the ceiling. Putting a sensor on the input and output of the air handlers gives us and idea of how much "work" the AC's are doing... and is the canary in the mine for AC problems. If the deltaT approaches zero and the room temp isn't below the set point of the thermostat, you have had an AC failure! We have "never" had an instance where the load on the room was zero.. so we have always needed cooling, even in the winter. We have had some "near zero" loads.. but it's been when we had a major machine room shutdown.. and even then the deltaT was always negative. With the "one sensor per rack" we spot overloaded racks. With the ceiling grid we spot areas where we need more "holes" in the floor.. and with the AC delta t we spot ac problems.... With a couple of hundred machines we only have about 40 racks. We can easily scan our entire collection in less than a minute.. and we log temps every 5 minutes. (RRD handles the average duties.) We alarm if we get three successive readings above a threshold temp... and will take either three readings from different sensors in the same scan or three readings from the "same" sensor on successive scans as sufficient to trigger an alarm. Three on one scan says the whole room is getting hot. Three on successive scans from the "same" sensor says that one rack is getting hot. Either way, it's an issue we need to look into.. and it has prevented "all" false alarms due to one hot reading on one sensor. (On the raw data, we would get a couple of "spikes" in a given month.. probably due to glitched readings.) Given that it takes about 40 minutes to get someone on site.. 15 minutes to trigger the alarm isn't a significant time period. (We don't have operators on duty 24 hours per day.) The biggest "step" we see in our sensor grid is when someone opens the back door to a rack, and cool air is mixed in with the exhaust air. (an open door is good for 7 to 15 degrees F depending on which rack.) Try instrumenting one rack "all over the place" and another just at the exhaust and see if you can convince yourself that you can see something of value in any given reading change before you commit to putting dozens of sensors in every rack. Another interesting experiment is putting a dozen sensors in a closed ice chest (no ice) and watching the results... There is a delta t, even in a fairly isothermal box.... which ones' right? Some of this may be due to self heating in the sensors.. (Yes, doing a reading does cause some energy to be expended the the chip.. more readings.. more heat.... this effect is probably worse on powered loops than parasitic power..) Our most useful reading for alarms is the deltaT on the airhandler.. it's where the first indication of an AC problem shows up. 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