Two and a half days in hell

Part one of Doomsday Weekend: who can you trust?

Sysadmin blog As sysadmins, we have to test before we deploy. We need to test before even upgrading a driver. We should test absolutely everything before a major deployment. It seems obvious. It is obvious.

You should certainly need to test everything before doing what I did: throwing ten times the normal I/O and processing load at your systems in a massive network upgrade.

But I am a systems administrator who did not have the resources to build a comprehensive test network. So this week I'll tell you the insights I gained from learning what not to do, and learning it the hard way.

Any systems administrator can tell you that major IT projects leave us with plenty of frustration to discuss: so welcome to the complete systems overhaul that I will call my Doomsday Weekend.

My IBM Model M keyboard has come to represent, for me, a happier time. It was manufactured five months before I was born. It has seen continuous usage since it was purchased; for well over two decades, ten hours a day almost every day. Every year I put it in the dishwasher and it emerges ready to go for another year. It's a metaphor for what I want to do: build technology, especially computer systems, with some semblance of permanence.

Most of the equipment I have to work with these days isn’t up to that level. My Wyse thin clients may outlive me, but they are a rare example of modern technology that isn’t designed to self-destruct the day after the manufacturer’s warranty is up. The built-in obsolescence of modern technology is really getting to me. During Doomsday Weekend, it nearly ruined everything.

The Doomsday Weekend has taught me many things. Some interesting technologies I hadn't explored, combined with punishing lessons of what not to do. None of those lessons has struck home quite as deeply as that of never, ever trusting a vendor 100 per cent. The number of hardware failures that we experienced during Doomsday Weekend was saddening.

An example is a five disk La Cie NAS that consistently tipped over after you put 2TB worth of data through it. A reboot solved the problem, but that is zero help when you are performing a network overhaul over a weekend, and the device won’t have available wetware to reboot it until the weekend is over.

Similarly, I have a smouldering hatred for pretty much everything ASUS has made in the last two years. I think the quality of its equipment has dropped, and the results have affected me directly. I don’t know from where ASUS sources fans, but I wish something bad would happen to whoever negotiated that contract.

Somewhere, there is a special hell reserved for DIMM manufacturers. DIMM failure rates are becoming ridiculous, and it doesn’t seem to matter which vendor of DIMM or motherboard we use. I’ve thrown everything I can think of at conditioning and testing the power into the systems, and I can’t find a problem that would affect anything. With six hundred DIMMs active in production servers, I see a DIMM die every two months. That’s nuts.

I could go on; I certainly would like to, but I am sure you get the idea. I am sure you all have your own horror stories.

The point I learned is: vendors can’t always be trusted. ®