Green Networking - Converting Standard Computers to Thin Clients

These days it’s all about saving money, power, the planet. This post is intended to get you excited to try thin client computing on your home network- assuming you have one.

My home network has for many years consisted of many sometimes-on Linux computers and one central server that does a huge variety of things from handling printing to operating security cameras or lights, to recording television programs with mythtv and providing an internet gateway/firewall. It may be on 24x7 but it earns its keep. This network is mostly connected with ethernet over cat-5 (100mbit/gigabit mixed).

So back on the savings subject. Hard drives consume a big chunk of the wattage going into a computer. They literally run little electric motors and servos and generate a lot of friction heat and noise too. They also cost more money when you’re building a new box, and they fill up and become obsolete fast. So if you’re building your own computers anyway, why bother?

Instead I use only one pair of .5 terabyte drives on the server computer. This system runs LVM2 and I can create new client partitions with ease out of this disk acreage. I have my DHCP server configured to feed specific mac addresses (belonging to netboot capable client computers) the appropriate configurations through pxeboot.

The clients when starting up query the central server for their boot information and are handed a kernel and initrd to boot. They boot up, mount their client partition over NFS and run exactly as they would if they had their own hard drive from that point on. Performance is impacted very little as ethernet is very fast and so are the .5TB hard drives on the server.

The end result is that I have a slew of client computers all running with nothing but ram (and sometimes CD drives) added to them, using much less power and generating far less heat and noise. The cost to performance is negligible, so it really is a win-win situation.

Some problems:

Grub netboot doesn’t support nearly as many network cards as pxeboot does. If you get strange errors or timeouts on the client, grub isn’t doing it for you, try pxeboot or just use pxeboot to start with. It is probably less familiar to you than grub but it is not hard to learn.

Compiling code over nfs is of course slower. Since I run Gentoo on several of the development clients, I have loaded them up with 4Gb of memory and mounted a tmpfs drive under /var/tmp of 3Gb. Now I compile in ram, and it is faster than a hard drive most of the time. Pretty much everything will compile in 3Gb with the exception of OpenOfficeorg. I install a binary of that.

shm /var/tmp tmpfs size=3G 0 0

Running anything that isn’t Linux may be tough to impossible. Install hard drives in the computers that will dual-boot, then instruct hdparm to shut the hard drives down when booted into Linux.

hdparm -S10 /dev/[hs]da

There is no (as far as I know) wifi netboot available as of this writing. So your notebooks, tablets, and other portable 802.11 devices will still need to use local storage to boot, or boot over the network on a hard line.

References

http://tldp.org/HOWTO/LVM-HOWTO/
http://syslinux.zytor.com/wiki/index.php/PXELINUX