by turing » Mon Mar 24, 2014 3:37 pm
Hi:
Sorry if this is too long a post, but I wanted to provide details on my experience. I received my cluster on Friday, three days ago; I checked the contents to make sure that I had everything that I needed and identified what documentation I'd need to read / view before starting. On Saturday I purchased a surge-protected power supply to use with the (supplied) power adapters and did a dry run of my procedure, using my Raspberry Pi 'cluster'.
On Sunday, I installed the current O/S image on an SD card, inserted it in one of the boards and, following the quick start instructions, successfully got it running - HDMI was fine, USB was fine [using an unpowered hub, as recommended, with just a keyboard and mouse attached], the Zynq wasn't particularly warm (I had, of course, put on the heatsink before doing anything else...) - so I proceeded to configure the SD card, installing the software that I am going to use, setting the timezone, password, et cetera. After a couple of hours, the Zynq was a bit warm but, otherwise, everything was normal.
I then cloned the SD card to the other cards and started testing. Each board passed, going all the way to the desktop after logging on.
I then began assembling the stack, as per the documentation. With each board, I attached the standoffs, installed an SD card, connected the USB hub and HDMI cable and then powered on the board - each started up fine. Note that I only powered on one board at a time, but the ethernet cables were attached to all the boards. All the power adapters were lined up the same way.
Finally, when all four boards were checked out, I connected the power adapters to all the boards and powered on the whole system.
I could not login to the board to which the USB hub was attached. In fact, neither the keyboard nor mouse would work, and swapping them with other keyboards or mice made no difference. Neither did using a different USB (unpowered) hub.
When I decided to try connecting the USB hub to a different board, I nearly burnt myself on the standoffs - not only were they scalding hot, but the boards were very warm, all the way to the edges and even the SD card was warm!
I then turned off the power bar - I had only powered the cluster for the amount of time that it takes to boot to the login screen - and let everything cool down before disassembling the stack.
I then tested each board - and none of the USB ports work now. The boards get to the login screen, and I can SSH into every board, but USB is 'toast' - which, considering how hot the system got, is not that far off. I had originally planned to operate the cluster as a 'headless' server, but now I'm very concerned about the potential fire hazard if I were to do so, since I'm not sure how big a fan it would take to cool the standoffs to just-warm-to-the-touch. If it takes a 20W fan to cool a 2W board, there's something very wrong going on.
Has anyone else experienced this?
What should I do next?