to reboot or not to reboot

[f. halili]

I came from a shop that reboots machines every week. I'm now working where systems are rebooted only during kernel updates, and must reboot.

What are your practices/recommendations ... do you have regular reboots? I'm working on rp7410's. - f. halili

[HGN]

Hi

This entirely depends on the you & the applications running on the server. There are some servers which are configured well stay and are never rebooted for almost a year but if there are some memory leaks you can get them rebooted atleast once a month or once in 2 months.

Rgds

HGN

[RAC_1]

I have observed uptimes upwards of 100-200 days. This is unix world, there is absolutely no need to reboot it every week, unless absolutely required. We require every 15 days, as we use clearcase heavily. Also a badly written apps, can go on eating memory and can cause performance problems. But that is just a bad apps, correct it.

Anil

[Rick Garland]

The rp7410's are good boxes. No need to reboot unless an application messes up. Example, a DBA screwed up doing a SQL script and the result was oracle taking lots of semaphores/shared memory. This in turned was filling up filesystems.

The reboot was required because of oracle - no other reason.

This is an exception. If all is well and no problems are detected, having an uptime in the hundreds of days is too be expected.

No need to reboot.

[Florian Heigl (new acc)]

We have different schedules:
highly productive: no scheduled reboots EVER, only once before mainteance windows to clarify the systems' status for the really risky work.
all others production/development: every week
integration/testing: when power fails :)

also, there are some forgotten systems, which have reached a stable state w/o application updates or anything neccessary or already faded out, those will easily reach 400+ days uptime, but of course the reason is that there is little reason to keep them up to date, or we just wait for someones OK to throw the switch.

[Kent Ostby]

The "reboot every week" mentality is left over from years ago when that was the norm for Unix due to a lack of resources.

As otherwise noted here, you may need to go this route if your applications start slowing down after a certain number of days up.

[Geoff Wild]

No need to reboot.

Only time I do is - patching, some kernel changes - and in in a few cases, application memory leaks - where the developers won't fix their code.

I have 7410's - great boxes - they will run and run and run....

Rgds...Geoff

[doug mielke]

we almost never reboot. So many people have to give permission, that we don't ask unless dire need.

We seldom have any problems.

BTW: There was a glitcth in old, non HP systems that limited uptime to 270 days or so, having to do with time being kept as a signed number which could be exceeded, changing the sign

[Rick Garland]

As can be seen from the previous posts, a regular reboot of rp7410 hardware is not required. These systems will run...

What you may have to deal with are some managers from the old school. They have been out of the trenches for some time and they are unaware of the improvements. As such their mentality may be to reboot because in the old days this could have been a good idea.

[Steven E. Protter]

We have a montlhy maintenance window which sometimes forces us to down Unix boxes for work on a SAN/disk array. To make it easy on operations, the policy is to boot during each of those windows.

Before we were SAN I would roll upgrades and patches through the systems and that resulted in a boot every 6-8 weeks.

There is no right answer. What works for you is what should be done. If there is no reason from the software standpoint, I don't have a problem with systems being left running 3-6 months.

HP-UX is a rock.

Solid. Dependable.

SEP

[steven Burgess_2]

Hi

We run a lot of middleware apps and have to schedule a reboot for each machine once a month. tuxedo is quite memory hungry and with some bad coding tends to run away with itself

apart from that, a lot of oracle back ends that have been up for months with no issues what so ever

HTH

Steve

[Gordon Morrison_1]

At least one flavour of Unix (HP-UX or TRU-64, I forget which) will lose track after 999 days of uptime, and won't know when it was last rebooted. That's how stable unix is (can be, when properly configured).
Compare this to the fact that after about 5 years, Microsoft issued a patch to fix a similar problem on Windoze 95 -- after it had been up for about 35 days! (It took them 5 years to get one of the b******s to stay up that long!)
A reboot can fix all sorts of problems, but unless there is a specific need to reboot (e.g. patch install, kernel rebuild, memory leak, zombie processes etc.) then why bother? 9 times out of 10, restarting the application/daemon that is having the problem will fix it. The regular reboot mentality is a hangover from working with less stable systems, and only increases your downtime (and workload). It should be resisted.
Reboot when necessary, by all means, but not just for the sake of it.

[Thierry Poels_1]

hi,

the only good reason to reboot a hpux server "regularly" is after applying patches (simply because your are forced to depending on the patches: kernel/software...)

Applications which consume and never release resources are another valid, but pitiful reason for regular reboot.

But otherwise keep the server running .. it can & it should

regards,
Thierry.

[f. halili]

thanks for all the inputs.