The Saga of the tg3 Module that Killed a Server (or The Monday that Wouldn’t End)

Our story begins in the early evening of an unassuming Monday evening, November 3rd to be exact.

After scheduling some downtime for a routine hard drive installation, our hero headed home for some much deserved time at home. Or so he thought (dum-dum-duuuum!).

The first crash was noticed at approx. 6:30 pm CST at which time our hero rushed up to find and fix the problem. All indications pointed to one of two students working on multithreading/multiprocessing programs for class, and their usernames were duly noted.

The second crash was a most unpleasant revelation at about 8:30 pm. Our hero noticed that only one of the two suspicious students was on at the time of the crash and consequently heaped all responsibility upon their unforunate head. After a quick reboot this user logged back in and out hero was able to observe their program at work. After about two hours of close monitoring and log combing our hero left for a few precious minutes of respite before his foray into the hard drive upgrade jungle.

The hard drive upgrade effort was a complete failure. The case manufacturers, in their infinite wisdom, decided to use a non-standard mounting scheme that requires hardware our hero just didn’t have. However, the server had stayed up reinforcing our admin’s belief that poor code had killed it.

Oh, but he couldn’t have been more wrong.

Upon finally returning to his house after 1 a.m., our hero attempted to salvage something of the evening by planning roadmaps for a few projects. At approx. 2:30 a.m. he happened to try to check his email from the newly unreliable server only to discover it’s recent demise. There was much cursing and grinding of teeth.

Our poor hero returned to his hell (common name: server room) only to find the network card installed in the server had joined it’s kind in the great network in the sky. After much more cursing, tho not quite as much grinding, and replacing the newly decesed card the server was brought to life for the fourth time that night. It was then that the current theory of student involvment was proven false.

Careful log perusal showed a metric ton of errors spewing from the network card’s driver for more than thirty minutes before the final crash. As the offending NIC had already been replaced, there was nothing for our hero to do but sit, wait and watch. By 4:30 am, the server had been back up for a little more than an hour with no problems so the dedicated admin was able to sneak a little sleep.

Arriving at work at 8:00 am, our hero was delighted to find all was well. The server was serving data quite efficently and, by lunchtime, the admin was able to return home for some much deserved sleep.

Leave a Reply