At AlliedModders we’ve been working on slowly replacing our aging “infrastructure” (big word for our little operation). In mid-2006 we got a Pentium 4 Xeon box at ThePlanet and the only operating system available was RHEL3.

Anyone who knows me knows I despise RHEL; yum is a huge pain. It doesn’t handle dependencies or core package upgrades well. It’s slow and its search is cumbersome. In order to not risk breaking the system we built everything from source and tried to completely avoid RPM packages. This meant a lot of cruft built up.

ThePlanet used to be good, but it seems their support quality has waned as of late (probably the result of the EV1 takeover). On two occasions our server simply rebooted with no explanation. They couldn’t provide answers as to why. In one of those cases, they simply closed the ticket saying “If you want to know, file another ticket.” Well, obviously, I wanted to know in the ticket already created for the incident.

Their control panel was decent but some information was just left blank — like serial access, which other big companies (like 1and1) provide. There was an incident where after 430 days of uptime, I got a call at midnight from CST from what sounded like a not-all-there ThePlanet technician. The conversation went like this:

TP: “I just got a ping notice that your server is down.”
Me: “Uh, well that’s strange. I didn’t turn it off. Why is it down?”
TP: “I dunno.”
(The delivery of this stupid statement dumbfounded me, so it took me a few seconds to recover.)
Me: “Well… can you find out?”
TP: “I guess… I can reboot it.”
Me: “That would be nice.”

After fifteen minutes it was still down, so I called the number they had called me from. This time they refused to talk to me since I couldn’t authorize myself (my customer information was on the server itself, a stupid mistake I won’t make again). I found this odd because they had called me fifteen minutes ago and taken a reboot order without even verifying my identity.

Eventually it got rebooted, but a few weeks later there was an explosion at their datacenter which took 9,000 servers offline (google “theplanet explosion” for coverage of the incident). We weren’t affected but it was the nail in the coffin – a few weeks later we completely moved to SoftLayer. I’d heard good recommendations of them in IRC, and they seem to be comprised of a few ThePlanet employees from times past.

(As an aside – I’m not the only one complaining about ThePlanet. My former employer had/has a large number of machines hosted at ThePlanet and frequently complained about support difficulties.)

Alongside that, we made a few big immediate changes. The first was ditching RedHat for Debian (my favorite distro because of its stability commitment, and the general convenience of aptitude). We also migrated to 64-bit. The next was ditching the “Cyrus” IMAP server. It seems no one can write an e-mail server that takes less than two days of work to configure properly. Cyrus was belligerent in refusing to consistently accept authentication methods and having archaic tools with poor documentation. Its configuration files were confusing and using its Sieve implementation required a weird FTP-like client. I’m sure that’s nice for a big setup but for <10 accounts it's a huge pain. So, Cyrus got dumped for Dovecot. So far I have plenty of praise for Dovecot. The documentation was pretty well written and had extensive use-cases with working examples. The authentication was easy to set up and its Sieve usage was trivial to get working. It even had an example of working with my favorite MTA, Postfix. I was able to convert over my 1.5GB maildir with only one annoyance (it didn't maintain whether I had read a message or not). The next big steps are moving to Mercurial over Subversion, and perhaps switching to Bugzilla over flyspray. Someday.