Is it time for a new server yet?

I’ve had my Linux server for several years now. I don’t remember exactly when I got it, but according to the time stamps on the picture gallery from when I did it, it’s been about 2.5 years since I improved the cooling with monster copper heat sinks (1 pound on each CPU). In my experience, a heavily used server like this isn’t good for more than about 3 years before it starts getting flakey. But so far it’s been solid as a rock. Since improving the cooling (and stopping running SETI@Home, unfortunately), I can’t think of a single time when it froze up or rebooted spontaneously.

So I’m thinking that although I don’t need to rush out and buy a replacement, I should at least start thinking of what to replace it with. And here’s the problem – computers have become way more powerful and fast since then. This computer was pretty fast for its day, and I could easily get CPUs that run 3 or more times the speed, but so what? There is nothing I do on this computer that stresses the CPUs in any meaningful way. Normally my load average is down around 0.01. So what would I want out of a replacement? “Bottom of the rung” CPUs, but really fast networking and disk? Something small and quiet like a Shuttle? Something maybe not top flight speed-wise, but really well built by a company that knows how to make reliable hardware like a Sun or IBM or an Apple G5?

One thing I really like about this computer, though, is the fact that it’s dual CPU. It seems to me that if one process runs away the other CPU keeps it pretty responsive until the process finishes or I figure out what’s wrong. For instance, yesterday I noticed the system getting pretty slow. “uptime” showed the load average up over 15, and “top” showed a process owned by the apache user called “oops” taking a bunch of time. One quick “/etc/init.d/httpd restart” later, and things were back to normal.

If I were to replace or improve this computer, I can only think of a few things I’d like to do:

  1. More RAM. 1Gb seemed like plenty when I got it, but since upgrading to Fedora Core 4 from Fedora Core 3, SpamAssassin seems to take WAY more memory so when I start up X and start doing stuff on the console (not very often) it actually starts dipping into swap.
  2. SATA. Right now IDE drives are wonderful and cheap, but it looks like the future is SATA.
  3. RAID. If I’m going SATA, I’d like to get a real RAID. I don’t know why, but it seems that most of the talk in the SATA world about RAID is RAID 0+1 (striping + mirroring), but I was really impressed the first time I saw a RAID 5 setup and the owner of it just yanked a drive out of the array and slapped another one in, and the application didn’t even hiccup while the RAID controller went about its business rebuilding the new drive.
  4. LVM. I like the fact that LVM can do a “transaction snapshot” almost like a database transaction, so you can backup a consistent view of the system instead of trying to copy an image of a system that’s changing while you copy it. I haven’t read if this is possible, but it seems to me that you’d be able to stop all the services that are most likely to have problems with consistency (postgres, mysql and innd for instance), start your backup snapshot, and then start those processes again, so the services would only be down for a few seconds rather than however many hours your backup took.
  5. Dual processors. Like I said, I consider that one of the best features of this current machine. Any replacement would also have to have them.

3 thoughts on “Is it time for a new server yet?”

  1. Instead of dual processors, would you consider dual core? You can get 1-series Opterons in a 939 pin form factor, meaning they fit into most consumer motherboards, and you don’t have to pay the premium for registered RAM.

    For real hardware SATA RAID, I’d recommend the Areca 1210 (4 port) or 1220 (8 port) PCIe controllers, since they’ve got a high-speed RAID processor onboard, and will fit in a PCIe 16x slot (either the main one and get a PCI video card or in the second slot on a SLI-capable board).

    Personally I couldn’t justify the cost of a hardware RAID controller, as Linux’s sofware RAID support (both 0+1 and 5)is plenty fast on recent processors. Hotswap isn’t quite as hands-off, but you only need 3 commands to replace a failed disk.

    What OS are you going to run? I regularly do installs of debian onto LVM on software RAID, with a standalone /boot.

  2. James, I currently run Fedora Core 4, but since I almost never use it as a desktop any more (it starts in runlevel 3 and I do a “startx” when I want it) I was thinking of going to Debian, just because it seems easier to install without X with Debian. Then I could maybe dual boot my GameOS box on the rare occassions when I need a Linux desktop.

  3. Debian is rock solid for servers, but be aware that if recent events are anything to go by (sarge was years late releasing), you might end up with an inconveniently out-of-date system and no simple solution.

    Sure, being a minor version behind with the kernel or bash isn’t too bad, but woody users were stuck with apache 1.3, php3 and exim 3 years after they were all-but obsolete.

    Obviously you can roll your own packages, and there’s always backports, but then you arguably lose a little of the stability that Debian’s famous for…

Comments are closed.