Server upgrade is coming soon

Due to the loading on the current dedicated server causing regular server crashes, I have decided to move CA and other hosted blogs to a new server with twice as much memory (2GB) and utilizing a proper server-oriented Linux rather than using a general purpose Linux. It demonstrates that if you’re going to do serious hosting, then you need to plan ahead rather than hope that your current benchtest system will do the job – it will, but not reliably long term.

At the moment, I am trying to decide between CentOS5 or SuSE Linux Enterprise Server.

What this will mean is that sometime in the next 2 weeks, CA will go offline for a few hours while the DNS updates around the world. I will let everyone know closer to the time when this is likely to occur.

15 Comments

  1. Steve Sadlov
    Posted May 29, 2007 at 9:38 AM | Permalink

    Since you (John A) are located – I believe – in the UK, I’d imagine that you’d get more similar time zone user community support with SuSE, given all the SuSE users in the Central European time zone, only one hour off from you.

  2. MrPete
    Posted May 29, 2007 at 11:42 AM | Permalink

    FWIW, I’d compare based on length of OS update support. I know CentOS, don’t know the answer for SuSE. (Fedora Core was a disaster along these lines!)

  3. Simen
    Posted May 29, 2007 at 12:21 PM | Permalink

    Only good things to say about CentOS – we’ve been running CentOS4 in our server environment for ~2 years now, and have not been bit by anything not resolvable (altho the resolving has been interleaved with swearing from time to time, of course).

    -S

  4. John A
    Posted May 29, 2007 at 1:28 PM | Permalink

    I’m used to SuSE Linux, but I want to see if CentOS can do as good a job with updates and fixes.

  5. John A
    Posted May 29, 2007 at 3:04 PM | Permalink

    The only catch with SLES is its not free to update beyond 90 days.

  6. Posted May 29, 2007 at 3:13 PM | Permalink

    They’re all good. If you want reliable support, Red Hat Enterprise Linux would be my choice. If price is the main issue, and your good with software problem solving, go with CentOS. It is R.H.E.L. without the paid-for support. I like SuSE, have toyed with it for many years, but the recent deal with Microsoft may have some customer implications down the road.

    Sonicfrog: Geology school drop-out, and Linux user and advocate since 2000.

  7. MrPete
    Posted May 29, 2007 at 3:47 PM | Permalink

    We’ve had two Centos servers running for about a year. Zero reboots, smooth and clean update/patches. And of course the price is right ;).

  8. Posted May 29, 2007 at 4:23 PM | Permalink

    I doubt the loading is causing the crashes unless your config is messed up – in which case your move won’t fix it.

    The most likely cause of random crashes, without a reason listed in the log, is a bad power supply, followed by failing fan, failed processor heat pads (use grease instead), flaky memory, then motherboard issues. Something like an Antec server class power-supply or a Tyan supplied case with supply and a Tyan motherboard – should help. We run Debian stable on Tyan motherboards (with registered memory) with out any problems. Some of the ‘dedicated servers’ supplied by hosting outfits are really marginal hardware. (Owning your own hardware can be a _really_ good idea.)

    IMHO Debian stable is just as stable as RH (if not more so) and gets security updates a bit faster.

    You should run memtest overnight on your new server with a bit of tissue over the air intake (to simulate a dirty filter to cause a bit of warming )

    There is a place in Los Alamos that will sell you server class hardware with Linux preloaded – they will burn in the system for you.

    Only reboots we have had with any Linux server (running mysql php apache) have been for upgrades – never had a web server crash from operations – now going on 7 years starting with our first RH server.

  9. Posted May 29, 2007 at 4:50 PM | Permalink

    Yeah, I shouldn’t have left Debian out of the loop since that’s what I’m using at this very moment. Apt-Get = Best package management system for Linux in my opinion. I suggested RHEL because John was wondering about CentOS, and since it’s RHEL without the Red Hat branding, I ran with it. Plus I didn’t want to start suggesting all the different Linux / Solaris / BSD options that are out there, and wind up with something not unlike a very familiar spaghetti graph…

    … throw everything against the wall and see what “sticks”!

  10. Posted May 29, 2007 at 4:52 PM | Permalink

    PS. “software problem solving” should have read “software / server administration”.

  11. Posted May 29, 2007 at 8:09 PM | Permalink

    Re: #9

    Seconded. I have used nothing but Debian in a very high volume production environment for the past seven years. We just run the stable distribution. When a new release comes out “apt-get update” and “apt-get upgrade” have just worked. I like being able to upgrade 600 servers 2000 miles away from my living room (and having it actually work).

  12. MrPete
    Posted May 29, 2007 at 8:20 PM | Permalink

    I’ll second Karl’s post, esp re memtest86.

    The most common point of “crasher” hardware failure we see is memory trouble.

    MemTest86 is your friend! We run an overnight test on every new system.

  13. Billy Oblivion
    Posted May 29, 2007 at 11:46 PM | Permalink

    I don’t know how ANYONE can say “Red Hat Linux” and “Reliable Support” in the same sentence without laughing.

  14. Posted May 30, 2007 at 2:17 AM | Permalink

    Re: 12

    Overheating is often a problem as well. Random segfaults for no apparent reason have often been the result of poor heatsink installation, gunk buildup in heatsink fins, fans slowing with age, and any other thing that can cause a CPU to heat up. Most of the modern CPUs will slow down their clock when they get hot so the result is that the CPU gets busy, work queues up, the CPU gets hot, slows down, and work then REALLY queues up.

    Moving parts such as fans and disk drives are still the main cause of failure for the systems I work with. Most people won’t notice a fan that has lost 25 to 50 percent of it’s original speed.

  15. John A
    Posted May 30, 2007 at 2:31 AM | Permalink

    The dedicated server is a rack mounted server in a datacenter, with an ERYC card to give me monitor/keyboard/mouse access.

    I did build it on OpenSuSE 10.0 for my own small purposes without envisaging that I would be hosting a popular scientific blog in the future, so I added a lot of convenient stuff on it for development and administration purposes, which hog memory and serve no useful purpose for hosting.

    Still tempted to go with SLES though – its what I’m most familiar with. I’m downloading CentOS5 to run in VMware to see if I can get the hang of doing the same administration tasks using YUM that I can with YaST.

Follow

Get every new post delivered to your Inbox.

Join 3,310 other followers

%d bloggers like this: