SysAdmin Adventures

Another Monday morning, another failed fibre GBIC in an Allied Telesyn switch. Yay(!). And so cheap at £80 apiece to replace. I wouldn’t mind so much if these switches worked well, but they really really don’t.

• Web interface sometimes works, sometimes doesn’t. If it can be arsed to provide the images (because ALL links are image links), it does so at its own discretion and provides no guarantee the right image will be in the right place.
• Switches enjoy failing completely at random intervals prompting a hard-reset.
• Sometimes, when they feel like challenging me, a single port will have a hissy-fit requiring that I reallocate the device plugged into the port, or hard-reset the whole switch.
• Upon reboot, the switch with the VLAN suddenly has a selective memory and forgets to put the VLAN back in place and so the segregated customer network suddenly has full internal network access.
• And, of course, there’s the GBICs, which die on average once every three months or so.

Thanks SO much Allied Telesyn,
No love.

Many manufacturers make low-quality hardware because it’s cheaper and they can sell it for less. Often, cheap hardware is just what you need; unless you’re buying UPGuards branded equipment. In this building we have a little bit of a power issue because instead of 230V we get between 240V and 255V depending on which way the wind is blowing. This may be the cause this may not be the cause, but the fact of the matter is, we have had somewhere in the region of 80 UPGuards PSUs, and I have replaced approximately 65-70 of them so far, and more die every month. Every time I have replaced one of these it has been with a bog-standard HEC 350W PSU and, let me be clear on this, not one of the HEC PSUs has failed on me yet. Not a single solitary one. These are not expensive or high quality PSUs, just cheap bog-standard quality.

If the HEC PSUs are fine, there is NO excuse for UPGuards to have nearly a 90% failure rate.

Buyer beware!

I understand that there are rare situations in which an initrd can be useful. For example, when hardware is constantly being swapped out, or when you absolutely must have an identical kernel image, or if you are using LUKS for your / partition. However, in general an initrd is, in my opinion, a completely pointless level of complexity that you are better off without.

I’m currently managing in excess of 15 completely different Slackware installs across lots of different hardware and not one of them is running a “huge” kernel, or an initrd. If a module is absolutely essential to your system’s ability to boot, why in hell haven’t you compiled it into your kernel? It’s like having a your starter-motor in your car disconnected from the battery, but having a relay on its own mini-battery connect it up when you turn the ignition.. why would you do it? What’s the point? The car will not start without it, why is it not hard wired into the system? It’s the same for the kernel. If the system won’t start without it, compile it into the kernel. Don’t play about with an injection system you don’t even need.

Ok, so your root device is inside RAID & LVM on a GPT partition. So, compile in mdadm support for your RAID level, LVM support and GPT partition support; job done.

For the less experienced, it’s also a good way to learn the basics of kernel compilation without needing to do much more than follow a standard process. All you need to do is use the config from the generic kernel you were going to boot anyway, and knowledge of what filesystem your root device is installed with and what storage controller your disk uses. Go into the kernel config, add them in [*], make, make modules_install, move and symlink the new kernel, update lilo, reboot. Once you’ve done it twice, it becomes so routine and easy you’ll wonder why you haven’t been doing it forever.

Then, once you’re more familiar with the process, you can start removing other parts of the kernel you don’t need, especially hardware controllers for hardware you don’t have, with each step making your kernel smaller and your system leaner and faster.

Have kernel, will compile.

Post followed-up by:
Compiling your own Slackware kernel