I stumbled upon this thread whilst googling "ubuntu server stability issues" - searching for answers to my own concerns regarding the stability of Ubuntu server.
I have to admit that I'm a long term Ubuntu fan, particularly on the desktop (Since Breezy).
Box 1: "Fred"
I first deployed Ubuntu server 8.04 on a production machine that has low usage requirements; it's predominantly a "brochure-ware" level webserver with about 4/5 websites, which also acts as an offsite backup repository. Primary packages are Apache2, Mysql, Postgresql, PHP.
It's dual core, has 2 GB RAM, 2x 1GB HDD configured with mdadm as RAID1.
Stability wise, it has been great except that it seems to die every 3-6 months for no obvious reason, despite combining through log after log.
I've kept this machine on 8.04, performing occasional updates.
Box 2: "Charlie"
Charlie has been running for a similar lifetime as Fred, and is used as an office based backup and media storage machine, office server monitoring node, network gateway for remote logins, wiki and virtualbox host.
Primary packages are: Apache2, postgresql, mysql, PHP, webmin, samba and Virtual box - Non OSE (We needed the headless feature back when that wasn't supported in the OSE).
Hardware wise, Charlie is Quad core, with 8GB RAM, has about 10TB of storage, distributed across a number of sata and ide drives, some of the sata drives comprise a soft RAID5 array, we have a drobo connected over firewire, two external usb drives and another drobo due to be attached.
Charlie started on Ubuntu 8.04, has been upgraded periodically via dist-upgrade and is currently at 10.04.
Sadly, Charlie is as stable as drunk in a brawl.
Charlie has frequent kernel panics, OOM's and requires a reboot every 2-3 weeks. Combing through logs has me scratching my head.
I love Ubuntu server, it's familiar, relatively well laid out, I love aptitude (Which should be the default package manager IMHO, packages/apps such as UFW, Fail2Ban, Denyhosts, logwatch, logrotate etc make administration relatively simple.
But both Ubuntu server boxes have uptimes measured in weeks or months, if we're lucky, and yes, during that time we've changed the hardware and re-installed from scracth, tested the disks, tested the RAM.
By comparison, I have clusters of HP DL360 G5's, DL380 G5's, DL380 G6's where uptime is measured in years, sometimes, 1000's of days.
These are running CentOS - and it doesn't float my boat like UBuntu Server, but it seems so much more stable, yet I don't know whether that's the Hardware or the OS.
Just my two-pence worth.