I like keeping my home workstations up as long as possible without a reboot. But, for a Linux box running production I feel it is better to give it what I call a sanity check by rebooting it on a regular schedule. Reboot every 30, 60 or 90 days.
The shop I'm woking in now has never done this. So that brought up the question, do you reboot your production servers on a scheduled interval? If so what's the interval? ...or do you think it is a dumb holdover type of idea from the days of older more troublesome hardware etc.
Thanks!

One of the selling points for linux in a production is that it can go for extremely long periods of times without rebooting, even years in some cases.

IMO the only reason to turn off a production server is to boot up into a new kernel or physically move the box to a new location.

There is only one reason to reboot a Linux machine - you are loading a new kernel.

Many people keep their Linux systems up for years.

That's a Windows mindset - "When in doubt, reboot!"

I agree it helps more often than not for Windoze, but this is a holdover from late eighties early ninties running AT&T 3b2's with System V os. Also a UNIX dirivitive named Gould made hardward and a custimized os which was a mix of System V and BSD that performed much better if we rebooted once a month.

Curently I'm running RHEL AS-3. We have big oracle databases housed on a SAN. We fracture (break the mirror) the DB every night for backup. Timing must be just right or all LUNS on the SAN do not remount. The only way we've fouond to get passed this is to reboot. After a reboot the nightly process works 80% better for a few weeks, then the problem comes up again. Thus the reason for my question. Thanks for the replies.

Most databases have the ability to unload the data to a flat file specifically for backup with integrity. I'm not familiar with Oracle, but I'd be amazed if it doesn't have this functionality. It also avoids the redundancy exposure and subsequent degradation from a split/rejoin for a mirror.

Another alternative is LVM2's snapshot capability which, again, is specifically designed for backups.

OMG! What a nightmare! There has got to be a better way.