High availablility and high uptimes are all very well, but what really matters is that the system is available when needed. For several years, I ran a system where a key application had a memory leak but only had to run from approximately 6:45 am to 11:15pm; we rebooted daily at about 11:15 pm and that application was stable. My uptime averaged 23:45, but we beat our application availability requirements.
I'm not impressed with 5-9's of uptime, if you get it by going to single user mode, doing maintainence and then booting back to run level 2 or 3. I'm not impressed with 5-9's of uptime, if you never patch, `cause you'd lose your uptime. I'm impressed with 5-9's of application availability, measured by your customer's business need for the application. Reboot all you want, just meet the business needs.
In the last year, I've seen several situations where a system has been having a problem and there is a known, elegant solution that will take approximately 20 minutes to implement and test before users can access the system again; or a reboot (15 minutes max) will fix the problem. Similarly, sometimes there is a 10 minute solution that might work, but if it doesn't you'll have to reboot - why not reboot right now, you know you'll be up in 15 minutes max.
jlb August 15, 2002