Yesterday I got into the elevator on the ground floor of building 14 on the Cisco San Jose campus. I parked my car and made my way to the elevator bank. The elevator arrived, I hit the #4 (for the floor that I work on), and the doors began to close. Right before they were completely shut, someone stuck their hand in, the doors opened (I was bummed out cause I had waited a while), and a person came in the elevator car with me. Then he hit the number 2. Now – obviously I could have taken the stairs up to the fourth floor, but it is a long way up… but one freaking flight of stairs??? I couldn’t help myself…”Dude,” I asked the guy looking really concerned, “Are your legs okay?””Yeah, they’re fine,” he answered.”So why didn’t you take the stairs?”He was incredulous…”That is the most insensitive thing I have ever had someone ask me,” he said as the doors opened and he left the elevator. This exchange got me to thinking about sensors, and what will be the best way to deploy them…There has been a lot of discussion regarding Wireless Sensor Networks (commonly called WSNs). The promise of sensors everywhere will spawn applications that will make the world an even more efficient place – I’m not going to write about some of these applications. Given my ‘insensitivity’, I thought I would write about what architecture WSNs might improve their overall sensitivity I.e. how well they communicate the information the sensors are recording.Like all networking technologies, we ultimately design around constraints. In the case of WSNs here are a few constraints:A) Most likely they will operate in unlicensed spectrum (unless a service provider decides to get into the managed WSN business).B) Many of the devices being sensed will be remote, and will not have access to power, therefore power is a concern. C) The sensors have to be cheap – especially if they are to proliferate. I once conducted a study and learned that people were willing to spend 1/15th the price of a device on an active RFID tag to track it. So if you want to track a 500 dollar item, then you better have a tag that costs <35 bucks. Now here are a couple of variables:1) Required throughput - sometimes this might be high, sometimes it will be a few bytes of information once a day. 2) Technology - Could be WIFI... could be Zigbee... could be the player to be named later...3) Business criticality - A reporting sensor on a heart rate monitor vs. a sensor on a remote laser printer definitely has different levels of criticality. So given the above... here is what I think...I know of two basic architectures for WSNs: Star and Mesh. A Star topology has each remote sensor connecting to a single hub - much like most Ethernet devices are connected to a switch. A mesh topology allows each WSN node to 'self organize' into a mesh. These nodes find a central hub (which may be multiple hops away. Therefore information a pressure sensor may take multiple hops through intermediate nodes before finding a central hub. Ultimately a WSN is deployed get information out of sensors. While the reliability requirements may vary from application to application, no one wants to invest in a wireless technology that performs spuriously. Therefore people should design for success and reliability rather than the cool factor. In my mind, knowing how complex mesh is, deploying a star topology is probably the right choice - especially because you will likely have time sensitive applications dispersed with less critical applications. I know that all of the sensor companies will talk about how well their mesh works, but is this technology for technology's sake? Are you willing to bet your WSN something that is neat and possible or something that is reliable? I'd love to hear what you think. Did this make me more sensitive?