Benefits of rebooting server

[Bruce Duff]

Hi folks, I remember reading somewhere that it is occasionally a good idea to reboot a HP9000 server to aid memory/performance etc. We enjoy superb uptime already and everything is great.

If anyone has any suggestions then please do drop me a line. I look forward to hearing from you. Thanks.

[Sanjay_6]

Hi Bruce,

There is no need to reboot a system is everything is working fine. Why mess with a perfectly good system.

Regards

[Bill McNAMARA_1]

To reclaim all memory in cases where you applications have memory leaks.

Possibilty to clear out temporary files in /var/tmp and /tmp automatically that can be difficult during runtime..

Apply patches..

But other than that, no real need.

[Rita C Workman]

There are folks who would agree with that philosophy.
But for UNIX, and myself, the only time I reboot a server is:
1)I have to replace hardware.
2)I have so many zombies and need to clean them up.
3)I have just changed the kernel
4)Some new high ranking mgr demands it......you know the type....

For #2 -
These are created when a process terminates improperly leaving child processes 'orphaned'. Thus they can generally be fixed by having the code corrected. So the sooner I can get that addressed, then #2 no longer occurs.
For #4 -
I wait for the next election, and those politically apptd mgrs are then replaced. Hopefully with someone less intrusive on systems they know nothing about.

I only boot a server when I absolutely have to. I am sure others have systems up longer...but I have servers up for months at a time without a reboot.....and they work just fine !

That's just my opinion though,
Regards,
Rita

[George_Dodds]

The only time when a reboot may be in order is if you have a number of zombie processes that you are unable to kill.

Cheers

George

[BFA6]

Hi Bruce,

I agree with Sanjay, if it is all working fine, then there is no need to reboot.

Tend to only reboot for hardware problems/replacement, patch installs and powerdowns

Hilary

[Helen French]

Hi Bruce,

There is no need of rebooting a server unless you have:

1)hardware upgrade
2)kernel tuning
3) Patch update which requires reboot ( kernel patches)
4) Application install/upgrade which requires reboot
5) Kill zombie process

There is no direct relation between server performance or reboot ( like windows servers do !). Read this thread which will have a lot of inputs for you from experts:

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0x7ab503bbece8d5118ff40090279cd0f9,00.html

HTH,
Shiju

[Andrew Kaplan]

If you're running in a mixed environment with servers running Windows, you're probably used to rebooting the Windows machines due to the periodic memory leaks that occur with NT. If that is the case, at that point if you want to do a clean sweep of all your servers you can include the 9000's in the reboot process. However, it is not something that I believe is absolutely necessary.

[Darrell Allen]

Hi Bruce,

That's "old school" philosophy. It may have been true in the early years but I don't believe it is now.

Poorly written apps can lead to system performance degradation, enough so that a reboot is needed to clear it up. However, the appropriate measure to take is to find out what is causing the problem and correct it.

I reboot after applying patches or making OS related changes. Many patches require a reboot. Many replace files that are busy at the time and are renamed. They aren't cleaned-up until a reboot.

Darrell

[harry d brown jr]

Bruce,

Rebooting a server, a unix server of any kind, just to reboot it, is senseless. I've had servers, again unix servers, that have run for years without rebooting, and the only reason they were rebooted, is because we moved them to another location.

Now lets talk stability. I have an N-class server, with six (6) virtual partitions that has been running since Oct 27th 2001, which is a little more than 4 months, and the only reason it was rebooted then, was so I could install the new release of Virtual Partitions.

If you want to schedule reboots, go for it, but use that time wisely!!!!! Use the time for patch installations and full system backups!!!


live free or die
harry

[Vincent Farrugia]

Hello,

Reboot ONLY if you have to. If you don't have to, you don't have to reboot. It is against the term "high availibility" big time.

When everything is working fine, say thank you to God and leave it running smoothly :-)

HTH,
Vince

[Bruce Duff]

Hi everyone,

Thanks very much for all your replies. I thought that might be the case but its good to get answers from fellow system people. Thanks again!!!

[Ian Dennison_1]

This will sound kind of rude, but I always try to get a reboot every couple of months, to "de-sensitise" the Users to the fact that they will not have the system available 25 hours a day, 8 days a week.

This is especially useful when management has promised the earth on availability times and failed to ask the techies if it can be delivered. It can make getting maintenance windows a lot easier.

Hope no-one else has encountered this problem!

Share and Enjoy! Ian

[Roger Baptiste]

If everything looks great, there is no need to reboot the system. That's the way we follow. There are systems out there which haven't been booted for months and some for over a year.

cheers
-raj

[Marcin Wicinski]

Hi,
there is no need to reboot HP9000 untill everything works fine.Some time ago I worked on a system with uptime of 308 days.
If you have applications which produces problems with memory leakage or zombie processes rebooting should help.

[A. Clay Stephenson]

Hi Bruce:

As the others have said there is no reason to reboot except for periodic maintenance (e.g. to apply patch bundles); however, before I apply the bundles (which have already been tested on a sandbox), I always shutdown and COLD boot the box. This allows me to find any failed disk drives BEFORE the patches are applied. It is not uncommon for a drive to fail to spin up and this is a good time to find them.

Food for thought, Clay

[Bruce Duff]

Hi there,

Well this question has put the cat amongst the pigeons hasnt it.
I was wondering what was the longest uptime anyone had enjoyed. I have managed a good 6 months before now (L2000 , now a diff. number) and only shut it down because of moving it to a different power supply. Thats almost as much uptime as a HP3000! (Now they really are bullet-proof boxes). Thanks all!

[harry d brown jr]

Bruce,

I've had a few servers that have been running, non-stop for over TWO years - those were the ones I had to shutdown and reboot because they physically needed to be moved to another area in the computer room - which was a shame, because that one system had more uptime than all of the NT's in the world combined ;-))


live free or die
harry

[Frank Slootweg]

One possible 'reason' which (I think) has not been mentioned, is to *test* if the system will (re-)boot correctly. I.e. sometimes one makes a mistake which prevents the system from booting, at least 'unattended'. It is best to find such problems during scheduled downtime, instead of after some kind of failure/problem.

On the other hand, one can make a new mistake *after* the test, so take this advice with a grain (pound? :-)) of salt.

Also, *before* doing such a test, make a recovery tape with make[_tape]_recovery(1M)!

[Bruce Duff]

Thanks for that reply. I think thats a good idea to have some scheduled downtime to apply new patch bundles etc. I, for one am terrible with syntax etc. and cant always remember how to do shutdown without refering to documentation etc. so its always good to do these things as refreshers on a regular basis.

Thanks everyone for your input, very helpful!

[support_5]

Hi,

Don't forget other reasons like testing that you can boot off a mirror disk (if you mirror the OS).

Generally though, I think it's good practice to aim to do everything you can to avoid re-booting your machine. This teaches you a lot about the system, and a little while later you can look at that little uptime counter increase and you will start to feel proud!

There is no performance benefit for regular reboots. But as someone said, there are user-expectation issues which could lead you to having a regular downtime...it doesn't even necessarily need to be a reboot though. Downtime and reboot mean different things but a manager might not know the difference. Thus, you could have a downtime (stop logins etc) but not have to reboot the machine. Even take it into single user mode and back if you had to.

I personally cant wait for dynamically loadable kernel modules to be available for everything. Then we can add and remove hardware, dynamically change kernel parameters, etc etc without needing any reboots. I think that's great!

I have worked on two machines that had an uptime of 1180 days!!! They were HP workstations too (715's). Only reason they went down was because they were being de-comissioned and sold off (old hardware).

Cheers,

- Andy Gray