Not the most compelling ‘stability’ argument
I've been running Linux for a long long time now (I think I flipped pell over to 0.12 or .99(something) sometime in fall 1992; I was certainly running it before I met the best in 1993) and one of the things I've always liked about it is that it's a pretty stable system on the crash recovery front. Pell (which, about 6 motherboards, cases, and hard disks ago, was the very first Linux box I set up) has been running pretty much continuously since 1992 with a grand total of one (1) case of data failure after a crash, and that happened when I overfilled the box with SCSI cards and it blew the power supply right in the middle of a news expire. I lost the inn history file, and had to regenerate it after I reshuffled the scsi busses.) And this is with the Linux ext2 filesystem, which, as all of the BSD fanatics will tell you, is Horribly! Unreliable! And! You! Run! The! Risk! Of! Data! Loss! If! You! Use! It! (an argument I found hard to believe to start with, and believed even less after the second or third freebsd crash which left me with a handful of files with good metadata, but containing nothing but pagefuls of ^A characters.) The other Linux machines I've run have been about as reliable, almost never dropping things on the floor when the local power company decided to play basketball with the mains power (these power failures taught me a few important things about the bsds, and those are:
- Not to use softupdates on filesystems that are important, unless you really like having to play disk recovery games to get back the contents of your /etc directory.
- Not to use either of the software disk array systems that a certain popular version of BSD uses, unless you really like to have heart palpitations at the thought that the SDAS de jour has just eaten 8 years of data when you see it pop up a syslog message claiming it's going to copy from the new hotswap disk over onto the old non-dead disk.)
But, be that as it may, the version of Linux I use is now old enough so that I can't just drop it onto new hardware and have it work. I bought a used Toshiba notebook computer a few months back that has an embedded wavelan card in it, and that card is new enough so that it's not properly supported by the (by now fairly ancient) driver that ships with the 2.0.28 kernel that Mastodon uses. So, since we're using RHEL3 at work, I decided that I'd get as close to RHEL3 as I could get without having to pay the boys in the research triangle a large wad of money; so I went out and got Centos, aka "RHEL without a large payment", and stuffed it onto the machine.
It has been, um, an interesting experience. Since I've put Centos on the laptop, it's crashed four times (two times because I disconnected the mains connection, and "used laptop" means "no battery left", once because firefox ran berserk and ate so much of the system that I couldn't get a C-A-BS through to the X server, and one mysterious "hey, it's tuesday! time to kernel panic") and, of those four crashes, two of them have lead to fsck eating the entire contents of /etc/pcmcia and, for good measure, wiping out the parts of /lib/modules that contain all the USB drivers.
And I'm running ext3 with journaling, too. And I want journalling, because I'd like to put Linux onto my main fileserver so I could set up a big old lvm and use filesystem snapshots to freeze my data so I'm not backing up live 60gb filesystems. But I certainly won't use it if a crash means I end up picking parts of a dead root filesystem. So I stick with BSD on my servers, even though the BSD filesystem support sucks and I have to keep a backup server running to do periodic rsyncs from the working server, because it's better than having to periodically reload the system from scratch and hope I don't miss anything when rebuilding it.
Sigh. And I really wanted to use lvm on linux.
Comments
I’m with Graydon. Laptop support has always been a problem with Linux, and you really need something bleeding edge, not a server-optimized distro like Centos.
I’ve been running Linux since the days of the 0.99.14 kernel, and so far I’ve never had any major data losses due to crashes, although I have had data losses due to hardware problems with an ancient SCSI drive. Plus there is a bug in inn prior to 2.4 where on a crash it would use the data that isn’t flushed to disk (the overviewdb) to correct the data that was flushed to disk (the history file), and that meant that the last crash I had I lost a bunch of news, but that’s not major. But when I rebooted to upgrade my machine from Fedora Core 3 to Fedora Core 4 last weekend, I had an uptime of over 160 days.
Of course my home server has a UPS. On a laptop, you expect the battery to act as the UPS, but evidently that’s not the case with old used ones.
I’ve been running a modified (heavily) redhat on my main server for several years. I’ve lost hardware three or four times, but the only data I lost was the one time I didn’t have a current backup. :)
On my home server and on my laptop, I’m running SuSE 10.0. Neither machine has even coughed in very long time. The home server gets accidently shut off from time to time (as in unplugged by two or four legged animals). The journal replays and all is well.
My biggest problem with most of the distributions of Linux is that the installer assumes you want a service running if you included it in the installation. I would rather it default to not starting up unless I explicitly call for it. I have to shut down many unwanted services each and ever time I update.
I do have two machines running BSD. They make up a two-machine firewall with a Red (internet) side, a Green (DMZ side inbetween the two) and a Blue (lan) side. I like the firewall tools that comes with BSD.
Well, it’s not very likely I’ll run redhat or any other commercial distribution on my file server; I’ll just brute-force Mastodon up to a more recent kernel/modutile/binutils/gcc/whateverthefuckelse Linus Torvalds now requires to build modern versions of Linux. What I’m worried about is the terminal fragilities of the filesystems, and that the way they go fragile is to go and stomp all over the directories two doors down from the directory that’s being modified. I could see /home/orc being lunched, because I constantly write into it. I could see /dev being lunched because every process and their sister updates the atime, but the only time the /etc/pcmcia directory gets tickled is when the machine boots up.
So it’s kind of offputting, particularly in the land of PGE and the amazing bouncing unreliable electrical network (and I can’t actually afford to buy a UPS these days between shovelling money away for emigrating and paying for the ridiculous copays and deductables that my “insurance” plan makes me pay for even the most piddly ailment.)
Comments are closed
RHEL isn’t really what I’d consider suitable for laptops; it’s got some support for them, but laptops as a species are twitchy beasts, and most of the work goes into the server hardware. (Best results are pretty widely held to be on Thinkpads or Fujitsu laptops if laptops are what you’re really after.)
I’ve been running various redhats and then Fedora since 97; no data loss, despite several dead drives, since I’m fairly religious about backups. Certainly no file system disasters. The Fey Creature lost data when both the source and the destination hard drives failed during rsync, but I don’t think that can be blamed on software.
Would you consider sticking FC4 on some non-laptop PC hardware for purposes of poking at it? I expect you’ll get better results.