Arch Linux

[Chess]

This discussion does bring up some interesting issues regarding updating Linux installs in general. I can't decide which is less error-prone: binary updates, like Debian/Ubuntu/Slackware/Arch, or source-based updates, like Gentoo/*BSDs.

I try to imagine what if I had a box installed with ABC distro in 2001, for example, when you had something like kde 2.2 and gnome 1.4 or whatever, and imagine then trying to update over the years all the way to present, with the latest and greatest of everything. I wonder which update method would have been more successful in the long run?

I have a Slackware box that has been updated using binary updates from 10.1 released in February, 2005 to v.11, released last fall, plus security updates afterwards. That's two years of upgrades without issue or breakage.

I had a Debian Sarge box that had an original Sarge install in June, 2005 with all updates to the present. That's almost 2 years of binary updates as well without breakage. That box has recently been replaced with FreeBSD.

My Arch workstation has been updated using binary updates, like I mentioned above, over the course of 1 to 1 1/2 years with some pretty major breakage. But, reinstalling from the ISO was never required.

I have read about people running FreeBSD, updating all along the way for many, many years, like 5+ years, without doing a "reinstall." My FreeBSD box is going to be a long-term test of how that upgrade process goes.

Anyway, I am rambling far off-topic. Sorry. I just am interested in the long-term viability of the various update paths out there and trying to figure out which one is more stable in the long run. It just seems to me that recompiling everything from source, ala the BSDs, although very time consuming, would generally work much better in the long run since you are essentially rebuilding and reinstalling the entire OS from scratch, just without using a ISO. It seems that trying to contually replace binary A with binary A.1, then A.2, etc. would eventually break.

[Wally Balljacker]

For the desktop the rolling-release is less effective yes, however with Linux distrobutions the average release cycle is every 6 months which is an aweful lot. To reinstall that often is a little bit much IMHO, which is why a rolling-release distro seems more viable to me.

I like a frozen distro for several reasons. They are generally much more stable than rolling-release distros by nature from my experience, and you don't ever have to worry about breaking your system in order to stay secure. In the case of Ubuntu, there are backports for many many applications, so being stuck with old software is simply not true. You don't have to reinstall every 6 months, if you take Ubuntu for example, LTS is supported for 3 years on the desktop, and 5 years on the server. You are in no way forced to upgrade, which is ideal for server usage. And for me, anyway, doing a clean install once or twice a year on my desktop seems reasonable.

[allix]

if I was running Yahoo or something similar, I would have a test box to test an update on to make sure it didn't break the application (PHP AJAX stuff just as an example) I was running on the server (You do that with all OSes on all production-critical servers), but once done, you install the patch and restart the server daemon you're running, and you're done.

Troels , you forgot to mention yahoo run a freebsd cluster

I have read about people running FreeBSD, updating all along the way for many, many years, like 5+ years, without doing a "reinstall." My FreeBSD box is going to be a long-term test of how that upgrade process goes.

So very true, FreeBSD has a reputation for clusters being online for years.
It would be interesting to see a cluster uptime statistics, and i don;t mean http://en.uptime-project.net/ or something similar because those are standalone machines which do not get upgraded kernels, which is stupid in itself, FreeBSD clusters on the other hand have new kernels without bringing down the cluster , you can upgrade the kernel on a different node offline and then connect it and take off the kernel on the other node, thats how it works as far as i know..

[CptnObvious999]

For the desktop the rolling-release is less effective yes, however with Linux distrobutions the average release cycle is every 6 months which is an aweful lot. To reinstall that often is a little bit much IMHO, which is why a rolling-release distro seems more viable to me.

I like a frozen distro for several reasons. They are generally much more stable than rolling-release distros by nature from my experience, and you don't ever have to worry about breaking your system in order to stay secure. In the case of Ubuntu, there are backports for many many applications, so being stuck with old software is simply not true. You don't have to reinstall every 6 months, if you take Ubuntu for example, LTS is supported for 3 years on the desktop, and 5 years on the server. You are in no way forced to upgrade, which is ideal for server usage. And for me, anyway, doing a clean install once or twice a year on my desktop seems reasonable.

Ok but you are mixing security updates with package updates, after a new version of a distro comes out new package releases become fewer and fewer as time progresses but you still get security updates for a while. Now when doing a world update on Gentoo I always get the latest and greatest in the branch forever which could lead to more breakage. But if I run glsa-check instead of doing a world update you will get a lot less breakage. So you get infinite security updates and maybe a little breakage with Gentoo for a lifetime or you could get security updates for another distro like Ubuntu which is less likely to break but after a while you no longer get updates. Which one is right for the system admin? Well there is no answer, I say Gentoo others say Ubuntu, it's really opinion based. All I am trying to say is there is a discreet difference between the way the two work that offer pros and cons and people should not look at everything the same.