We were going to move all the servers from our old ISP to a new one, and in order to prepare a bit for all eventualities, we moved a single server on Thursday. It was a good experience, and we discovered stuff like the fact that the network was broken. Or rather, it wasn’t up yet. One of the people from the network provider was there and we chatted a bit before deciding on we would leave the server in the rack and he’d call me if he had any problems getting the network up and running. It was up sometime on Friday, so that was nice.

Friday, we ran about finding missing pieces we needed for the move. In addition, I got some more registrar work done. An SSL certificate was ordered but it got hung up somewhere in Thawte’s system.

Monday, we ran even more about, this time actually finding the parts we needed. The servers were shut down at 2200, we finished getting them out of the rack and into the trunk of Amund’s car a little after 2400. Getting the servers into the new rack didn’t take too much time, but configuring everything correctly took a while. We had some issues with Apache’s virtual hosts and DNS, but it got worked out. I also managed to misconfigure the mail server (or rather, its DNS entry) somewhat, so the route ended up somewhere in Italy. We finished at about 0730, and even though everything wasn’t triple checked, we were so tired that going home was the best option.

As I came into work at 1300 today, Tuesday, I got a pile of stuff thrown onto me: the mail server didn’t work properly, a few hosts were misconfigured and so on. Fixed this and sent in the two change requests to NORID so we ended up with three DNS servers operational again.

Had some food with Anders and Amund at Peppes before fixing tying up some loose threads. Ended up discussing my fan control project with Amund before going home.

Am I the only one who is annoyed at the late, broken posts at Planet Debian ?

Please people: fix your feeds.

GUADEC is over and I’m currently sitting on the train back home. It’s been a good conference, meeting new and old people. The talks were ok-ish, but I realize that the real reason I go to conferences is the people. Chatting about both technical and non-technical issues is something I learn a lot from. In addition to the DDs there, the Skolelinux people, random GNOME/Ximian/Novell/RedHat hackers, I had the pleasure of meeting Bdale’s family.

At conferences, I have a tendency to run into Matthew Garrett, and I did this time as well. He helped me get ACPI working again, even though we didn’tmange to fix it properly. (Using a couple of patches off “the kernel bugzilla”:bugzilla.kernel.org.) I fiddled around with it on the train and it seems the ACPI controller doesn’t like to have its interrupt method (level or edge) fiddled with while restoring, so I just added it to the mask of reserved interrupts. It now seems to work just fine, suspend - resume - suspend works, and I think my network card works after resume. Yay, yay, yay!

Most of today was spent helping webdeal with one of their Opteron servers. They are more used to Intel hardware and *BSD, while said machine runs Linux. They have had some performance problems with another one, so I helped them tune this one a bit, and it ended up with fairly nice performance.

After some food, I went to the server room and continued to work on the server, now reinstalling it with Debian (we had done the tests in a Debian chroot inside a Mandrake installation, since that was the CD I had available). Unfortunately, I had to cross-install using the already existing installation as the Woody CD didn’t find the RAID controller. It worked fairly well, but during this process the whole circuit breaker decided that the rack had overloaded it, so it threw itself and the whole rack went quite silent quite fast. After a few phone calls and running about, I managed to get it all running again, just a few systems needed a manual fsck.

Got the system installed and left the server room, tired but happy.

A rack losing its power and then all the UPSes failing makes a hell of a lot of noise before becoming very, very quiet. And then it makes a lot of noise when you make the power come back.

Note to others: Do not, I repeat, do not attach a full rack of servers to one 10 amp circuit, and do not attach the UPS-es to each other so you’ll be sure to overload the whole shebang.

Yeah, and fix stuff, don’t document workarounds when the fix is easy.

This hasn’t been the most productive of days, I’ve been doing a little registrar work and cleanups, but not very much. A bit tired after last night. Apart from the registrar work, I’ve written a draft requirements spec for a project, which will be quite interesting once it goes live.

Finally got around to signing all the keys from Debconf 4. I ran into a couple of problems, some related to the tool I’m using, gpgmailsign (which, after what I’ve understood, works fairly similarly to Keybuk’s signkey ), some related to the fact that some people don’t have any encryption keys connected to their key at all, which is in theory fine, but I really don’t want to diverge from my usual way of signing keys. Another problem which popped up was GPG messing up (no valid OpenPGP data found), missing key on keyserver, etc. If you’re still missing my signature from Debconf, drop me a mail and I’ll see what I can do.

Today was just spent doing registrar work. Tedious work where you have to get it exactly right, else you get the application refused (though, with an explanation). All the domains we administer should now be in order, but I believe I’ll need at least another day to clean out the last small problems. We are going to change ISP in a few weeks, which means lots of work if the DNS isn’t set up properly. (Or rather, it means even more lots of work if the DNS isn’t set up properly.)

As always, we discussed around a bit and I begun helping out another guy with a broken file system, but I didn’t get around to actually fixing it, so I’ll have to look into that tomorrow or so.

Friday, I managed to actually get some work done. Anders brought back one of the servers from the server room on Thursday, and the reason for it being broken was one of the coolers having come loose. The cpu was fairly burnt, and after a short while, I gave up getting it to work. Left work early to meet up with Karianne.

Monday, I’ve been mostly in a board meeting. We discussed for some hours, slowly getting through the list of items we were to process. Due to an oversight, we had to cut the meeting short, so we’ll continue on Wednesday. Will be fun, I guess..

The rest of the day, I spent doing registrar stuff, going through the huge pile of mess that is our current set of domains. It’ll be good once it’s cleaned up, but right now, it’s a huge pile of work, just waiting for me to go through it tomorrow.

Manoj writes a bit about [appliances versus tools](http://www.golden-gryphon.com/blog/manoj/software/manoj.2004.06.17.golden_gryphon]. He’s saying what I’ve been trying to say about user interfaces for a long time: A bit simplified, you can divide the “cost” of using an UI into two parts: One initial cost, which is what it takes to learn the interfaces, and one running cost, which is how much it costs to use an interface after you’ve learnt it. Of course, this is a bit simplified, since learning a tool is not binary, but I still think the comparison holds.

For tools which you use seldom, you care a lot about the initial cost, since you will incur that many times as you have to re-learn the user interfaces. An example would be an ATM. For a complex tool, such as my use of a computer, the running cost is what matters. If I have to spend a week to learn a tool I’m going to use a lot, it’s worth it.

However, as Manoj points out, this is not only a result of complexity: Both ATMs and general purpose computers are complex beasts, and an axe is a lot simpler than a refrigerator (mechanically speaking). The HCI term “affordance” doesn’t quite cut it either, as it is more about how easy an interface is to learn. On the other hand, an axe affords a lot of things, that is, it doesn’t give you much guidance on how to use it. A refrigerator door affords opening, it’s one of the few things you can do with it.