Regular Rebooting Servers (RHEL5 vs Windows)

[SteveC]

Hi all,

In the past we've regularly rebooted our application servers (Win2k and Win2k3) every week or even every night to ensure a clean environment for critical apps/DBs. Now we're moving to RHEL5 is this still a good idea or is it even detrimental?

Cheers,
Steve

[cjolley]

We went to Linux for our Oracle database server last year.
It runs Oracle Linux, but that's just RH with some artwork changed.
We thought we could get away from our weekly reboots once we got on Linux.
No way.
Linux got even weirder than Windows after about a month.
So it's back to once a week reboots, which is really not a big deal to us.
The Linux version of Oracle DB EE runs much faster than the windows version though.

[Gurm]

Uh... I haven't rebooted a Win2k3 server out-of-cycle in... honestly, years now.

Now there are monthly patches, admittedly.

But some of the servers have run continuously for a couple years without a reboot. No problems to be found.

Linux, on the other hand...

[Claymonkey]

It really depends on what its doing. Citrix servers... you should reboot them weekly. DB servers it depends on the db. VMware servers... probably no since all it does is take and give memory and resources to the virtual servers... which from what I've seen play nice. As for the normal file/print server. I can't tell you how much better Linux cleans up after itself than a windows server.

So... its a big maybe. review what your server's roles are and research if they have memory leaks or if they gracefully handle a program crash. Also, linux auto-cleans the tmp directories on reboot. If you aren't rebooting have you configured it to auto clean them on a regular basis? Some distros do, some don't out of the box.

[Jammrock]

We only reboot for patching and service issues (such as terminal services dying ... again). We're a pure MS shop, except our VMware ESX servers which are VMware Linux.

[MultimediaMan]

It's a rare day we boot a Windows Server for anything these days, out-of-cycle. Even with all of the MS Updates, we don't reboot more than once a month... this month's updates (05/08) did not require a reboot.

Anytime we reboot part of a Windows Cluster it is a major deal; we have a playbook to follow: customer notifications, etc...

Now, we have been very hard on our software and hardware vendors to give us gear that is up to snuff: We capture dumps on every software failure we see now as a matter of policy. It really does help narrowing down where the problems are - and most importantly, how to get them fixed.

Quite a few times we find that otherwise unexplained software problems can be a predictor of hardware failure: a minidump can usually pinpoint the problem when we "play it back". Most of the times it just finds a flaw in the software - dump analysis is still more art than science - more experience using the tools against a piece of code usually results in quicker resolution: some software is easy to debug, others can be a real nightmare.

[UtwigMU]

I have to reboot SBS boxes about once a month otherwise they run over RAM (1-2GB for older ones) and services start failing. I have only one Debian FTP server and I just patch it without rebooting.

[MultimediaMan]

Remember this...

... we have been very hard on our software and hardware vendors to give us gear that is up to snuff: We capture dumps on every software failure we see now as a matter of policy. It really does help narrowing down where the problems are - and most importantly, how to get them fixed.

As you revisit this...

I have to reboot SBS boxes about once a month otherwise they run over RAM (1-2GB for older ones) and services start failing. I have only one Debian FTP server and I just patch it without rebooting.

If you are rebooting SBS for memory leaks, you have what I would consider major software problems. It never hurts to hold your vendor's accountable.

[SteveC]

Well, looks like we will be rebooting the new RHEL5 environments nightly! The vendor's software has proven to fail to shutdown itself (or even be manually killed) regularly and it takes a reboot to clear. Just like Windows then

[Gurm]

I have to reboot SBS boxes about once a month otherwise they run over RAM (1-2GB for older ones) and services start failing. I have only one Debian FTP server and I just patch it without rebooting.

SBS is an evil thing. It's attractive to small businesses because of MS's licensing model, but people routinely run it on 1-2GB servers when it's running:

- Active Directory
- Exchange
- Fileshares
- Printer Sharing
- SQL
- Backup

ALL AT THE SAME TIME. And you wonder why it runs slowly/poorly!

However... that said... it shouldn't be having memory leaks. That's a problem. Fix it.

[UtwigMU]

Well, I think it's mainly due to McAfee, Groupshield and Backup Exec + proprietary accounting app tacked on top of that on 1-2GB boxes.

[MultimediaMan]

Narrow it down: look at pagefaults, handles and committed RAM... you should be able to find the culprit.

It's probably not McAfee: as much as I dislike their products, It's what I am most familiar with troubleshooting professionally.

Not familiar with the other two...but anytime you say "custom or proprietary" I shiver...

[Gurm]

McAfee can't run on the same server as Exchange. Period. It's a known bug they're still working on as of a month ago.

[MultimediaMan]

I will disagree with that statement; I will qualify that we do have McAfee Engineers in the building. And we have run McAfee and Exchange for over 9 years...

[Jammrock]

Isn't McAfee Engineer an oxymoron