Down For Maintenance Sometime This Weekend

When I set up this blog four and a half years ago, the performance of some of the journaled filesystems that did not originate from Linux, like JFS and XFS, were better than the native filesystems, notably ext3, which was really a journaled version of ext2. So the blog, for a number of years, has run off JFS, which had its origins in IBM’s flavor of Unix, AIX.

But from most benchmarks I’ve been able to find, and my own testing, ext4 is better than either JFS or XFS in terms of performance. The blog is running on a very old filesystem, and it’s time to make the switch. The defrag we’ll get from the switch alone should boost performance, even absent the filesystem switch.

I estimate the blog should not be down more than an hour sometime either after 1AM Saturday morning or 1AM Sunday morning. One of the great things about being linked by Instapundit is it gives me a great opportunity to examine the server under a heavy load, and I’ve come to the conclusion that the filesystem we’re using is both badly fragmented, and using a slower journaled filesystem than we could be using.

14 thoughts on “Down For Maintenance Sometime This Weekend”

  1. Best wishes for the changeover.

    I don’t think I’ve ever done a large-scale filesystem migration like that. Definitely not on a public-facing server.

  2. It’s not really large scale… the blog only takes up about 2 gigs. If I were really crafty I could do it so there would be no downtime, but I don’t care that much :) Doing it offline saves me a lot of work.

  3. I should say all the blogs I run, which is about 6, only take up 2 gigs… this one itself is probably under a gig.

  4. Now that you clarify it, it doesn’t sound large-scale.

    But in terms of the number of people who might email you if the thing goes belly-up, then it may be large-scale…

  5. Cool, good luck with the switch. I really should hook up an older machine and try playing with Linux sometime, never got into it.

  6. I read the post twice and I still can’t figure out what the heck you were talking about…. My geek card has been shredded!

  7. Have you tried UFS from the FreeBSD line? I’m not up to date on my filesystems or their performance, but curious if you had an opportunity to try it.

  8. UFS isn’t a very well defined standard, as there have been extensions added, depending on the vendor, and not well supported in Linux. It’s not really an option.

  9. Well I hope you get your EXT4 in place of your JFS or XFS PDQ… I B checking back…

    Dann in Ohio

  10. If the blogs only take up 2 gigs, it shouldn’t even be hitting disk to serve pages if you’re doing things right.

    All demanded content ought to be cached in RAM…

    If the filesystem is your chokepoint, there’s a design problem in the blog platform, with that small of a content base.

  11. I had a few things configured that were less than ideal, but JFS was definitely a problem, as switching to ext4 seems to have helped quite a bit. It’s barely touching disk at current traffic levels. We’ll see how it does with a Glenn Reynolds link, hopefully soon. I suspect it’ll do fine.

    This isn’t a machine I had done a whole lot of optimization on, because under normal traffic, it holds up just fine.

  12. All demanded content ought to be cached in RAM

    Normally, it is. When a Glenn link comes in, it forks more apache instances, which leaves less room for disk cache. I have 8GB of RAM on this machine. I should probably have more, but it’s only a Core2Quad. If I went to 16GB, which probably would be better, I’d prefer to use ECC memory.

Comments are closed.