A belated 2010 follow-up to the last year's experience. This year, as you've probably seen in the news, was quite fun. Thanks to the help from BELNET (and of course the Fosdem folks who did all the discussions), and the two rolls of fiber, we had 1gbps uplink, and 1gbps link between H and AW.
The most stressing moment was when on Friday at 5pm we discovered that 5 out of 6 strands of fiber that were planned to connect us to the internet, were not working. Plan B (which plan B? we already tested everything! it worked!) was pretty weak. So, applying some logic and with a bit of luck, the problem was solved by resplicing one of the ends.
After that, a few hours trying to build the correct OS platform to install the management software, which supposedly could magically convert some of the APs that we had into the correct mode, rather than us doing manually. At something like 11pm we realised manual was faster - but we didn't have the power supplies with us, but since the security folks wanted to close the doors, we grabbed a few access-points and went to the office to fix them up.
The process of converting access-points from standalone into management mode is dead-simple - hold the button while the AP is powercycled, then power it on with the button is pressed, and wait till the two out of three LEDs on the top turn red. Then release the button, and the AP will grab the predefined address 10.0.0.1 and will start the TFTP download of the image with the default name. All good except the TFTP requests would be sent to broadcast address. It took me about 7 access-points (as measured by Jerome converting them with some naive TFTP server under windows), till I have figured out how to get the atftpd to react: "ifconfig eth0 promisc". Really-really weird. I thought tcpdump - which was running in parallel - was already putting the interface into promiscuous mode.
Anyway, once I found the trick, the rest of the conversion was trivial, so when I started up at 5am, the remaining of the APs were trivial and fast to convert.
This year the access-points we used had a bit more juice - both on 2.4Ghz and 5Ghz - so we managed to kick off some interference to the wireless mic in Jansson. And this year we used the feature to "kill" the rogue APs - which were apparently not so low in number. Strange. I thought people would come to the conference to listen, rather than to annoy others. Nonetheless, killing the phy seems to work - at least this year I saw much less of "strange" SSIDs. Next year we should enable the triangulation, to go and ask personally :)
The kicker this year was that the network was full-dualstack. And, I cooked the bind9 to point to the SixXS recursive servers (thanks guys for the service!), DNS queries going over the SixXS tunnel. A bit of a chewing-gum and matches, but worked fine over the period of two days - and looking at the SixXS traffic graph, it felt like we even made some bump for the gateway by our queries - although this is most probably wishful thinking.
I don't know how many people noticed their google.com traffic was IPv6-based (being geek conference and such). But if they did not - even better! That means - IPv6 works!
All in all, it seems like everyone was quite happy - the visitors (thanks for being there!), the Fosdem staff (it was fun, guys! Let's do it next year!), and ourselves.
On to FOSDEM2011!