Hello,

Please can anyone tell me - is it true that you should not re-boot Unix Sun Solaris servers on a regular basis, but onyl do it when really required?

We want to schedule a reboot on a daily basis, to clear any rogue processes, but have been told this is not a good idea.

Can anyone advise?

Thanks,
Paula

Indeed, you do not boot Unix systems unless there is a valid reason to do so.

Unix isn't Windows which messes up memory etc. all the time.

And booting a system just to kill processes isn't a very good reason. Processes should terminate in a "normal" way. If they don't you better talk to the developer and kill them manually when ever the need occurs.

My policy, and this is one man's opinion, is that a Unix box should be rebooted every 4 months or better. Too many times I have seen a Unix box up for 8 or 9 months that would not reboot cleanly because of a hardware problem or a startup script problem. The scheduled reboot exposes this stuff. Also, at my last job, we had a critical Sun system that had been up over a year, with /.reconfigure present. The admin who put it there apparently had left the company and the rest of us did not know what to expect. (the answer: path-to_inst file was corrupt...) When a lightning storm took out our UPS, we got to discover stuff like this for 600 servers.

Another issue is that patches usually require reboots and a box that is setting records for continuous uptime is pretty much guaranteed to be missing some patches. Still another is that complex clusters sometimes require special boot sequences and these can change as services are added to a cluster. Cross dependencies between clusters can further complicate this.

On the other hand, daily or weekly rebooting is silly. I could tolerate once a month but I would try to push for 60 days. So my range is roughly 60 - 120 days and I'm willing to accept 30 - 60. (But I did not make the policy and we kept the boxes up until we were forced to reboot.)

I agree with Perderabo....it is all based on requirement of course.

If the need is there, then it has to be done, but 4 months - 6 months is a decent average time frame. However I have noticed that most reboots are done because patches need to go on or more external disk needs to be configured.

I know for sure in certain financial firms, they reboot boxes every week....over 10,000 a week. so it really is dependent on environment.

-S

I'm working on rolling our the lastest updates for Solaris and AIX to address the daylight savings time changes. In our environment we don't reboot on a schedule. Many machines have been up 200+ days, and probably 20% or more have been up over 400 days. We've even seen one 900+ and one 1500+ uptime.

There is a very strong correlation between how long the box has been up and how likely it is something will go wrong with our update process or the reboot at the end of it. If a server has been up for hundreds of days it will be downlevel on many more patches (and farther downlevel at that). Additionally there's more time for minor problems that go unnoticed in day-to-day use to build up. Between those two factors, you are really playing with fire leaving a box up forever.

Of all the places I worked my favorite policy was one where database servers were rebooted every 90 days, everything else 120 days. That seemed to be a good balance point where you weren't having too much extra work but still had fairly fresh servers.

We try to go with ~90 days. We take patch updates once per quarter ( unless a critical issue arises ) and reboot as part of the procedure.