Reboot HP to refresh performance?

[Jim Shepherd_1]

We have an HP9000 server running HP-UX B.11.00. Our system administrator has just requested that we free up the server because it has not been rebooted in a couple of months, and he wants to "refresh" the server's performance. I have administered SGI servers in the past and have never heard of rebooting a server for performance reasons when everything is working fine. The server only runs one application, ABAQUS, which is a finite element solver. I have a feeling our "admin" has a Windows background and is just using that experience for the request. Is there any benefit in rebooting an HP-UX running server when everything seems to be working fine? It is a hassle to have to get all of our users to stop running jobs at the same time for this seemingly worthless proceedure. Thanks for any comments.

[A. Clay Stephenson]

Get a new admin. My first reaction is that these guy is seeing a lot of zombie processes that he can't kill with kill -9 and it's driving him nuts! There are a few cases where a reboot is need to 'refresh' performance but that is fixing the sypmtoms rather than fixing the problem; the most typical of these is applications with memory leaks. The other case might be to cleared shared memory because you don't know which processes have it attached. In general, the only time a UNIX box needs to come down is for scheduled maintenance.

[G. Vrijhoeven]

Hi,

No, if the aplication is stable and cleans up its processes (No zombies). I do not see a reason to reboot the server. He must have a reason for this request. Ask for it.

Hope this will help.

Gideon

[James R. Ferguson]

Hi Jim:

This makes no sense unless you have an application that has memory leaks (and that's an application problem that needs fixing) or you have hung ports (another application problem, really).

Reboots should only be necessary for kernel changes, some patches, and hardware upgrades.

Regards!

...JRF...

[Jim Shepherd_1]

There are no zombie processes running on the machine. The only application we use only runs for several hours (sometimes a day) before it ends on its own, so it doesn't seem likely that a memory leak would effect long term performance. I plan to tell the admin not to reboot the server until something breaks, but I wasn't sure if there was any substance behind the admin's request.

[Victor BERRIDGE]

Well,
Good reasons for regular reboot is :
Reinitialise all the logs
Clean /tmp
Check on file systems
Get rid of zombies etc...

Now there are cases where for production purpose you cannot do so (We have 2 boxes like that...)and so these boxes only get rebooted when you add some new hardware etc...
AND they work well (...)
even installing patches is not allwed unless you can proove the box IS in danger...
So now up to you to see if you can find a justification, it is true that if all works well than it can wait until more suitable period...

All the best

Victor

[S.K. Chan]

Clay is right on. The number one reason for a constant reboot is when you got a memory leak situation. The only way to deallocated the memory back is by rebooting. Otherwise over time all its memory will be used up. A good syadmin regularly monitor the system resources to make sure there isn't any gradually degradation to system resources.

[Marco Paganini]

Hello Jim,

Looks like you have a Windows administrator working there. :) Tell him that this is U-N-I-X we're talking about. One doesn't need to go around rebooting the system. The only exceptions are when you have zombies, processes that won't die (locked inside a system call, normally because of a failed device, etc).

We have HP boxes here with 940 days uptime working like a dog and not a glitch.

Regards,
Paga

[fg_1]

Unless you are running an Instance of oracle that is doing a significant amount of Write's vs read's then I would not worry about this.

There are some applications out there that cause memory leak errors such as sybase or informix, and to a lesser degree oracle, with these database apps then you would effectively want to reboot monthly.

Although I must admit in speaking to alot of other admins at event like hpworld and interex, it's becoming a more common practice to reboot servers at least monthly.
(bad administration practices cause bad systems)

[Jim Shepherd_1]

What is the best way to check for a memory leaking application on HPUX? On IRIX, I would use osview to look at the memory usuage. Is there a similar utility on the HP?

[Thierry Poels_1]

Hi,

just repeating others : "reboot rarely required; only with memory leaks and/or hanging processes; or off course system maintenance"

If the guy wants a reboot, let him install the latest HP-UX patches: he'll get his reboot, and you'll have your system uptodate ;-)
If he was smart he would have asked for it himself; it's a nice way to maskerade a reboot ;-)

regards,
Thierry.

[Steven Gillard_2]

Ummm, even if you do have a process with a memory leak, a reboot shouldn't be necessary. Just restart the application. Likewise for zombie processes - restart the application that is leaving them undead (ie kill the parent of the zombies) and it will start with a clean slate.

Having said that proactive reboots can be a useful thing, especially after you've installed some new applications and want to make sure that everything starts correctly when the system boots.

But a reboot to "refresh" performance is a waste of time. If everything is running ok then leave it alone.

Just my $0.02. Regards,
Steve

[S.K. Chan]

hi jim,
attached would help in checking for memory leak.

[James R. Ferguson]

Hi (again) Jim:

To answer your question about detecting memory leaks, do this:

# UNIX95= ps -e -o ruser,vsz,pid,args|sort -k2nr|more

You will need to watch for growth in the number of virtual pages of memory used.

Note that the UNIX95 (XPG4) option is armed only for the 'ps' command. There is a blank after the equal sign and no other delimiter before the 'ps' command.

Regards!

...JRF...

[Christopher Caldwell]

A reboot may "seem" to "refresh" performance if the file system buffer cache is competing with applications for physical memory.

If this is your case, it's probably a better idea to correctly tune the buffer cache than to reboot every few days/weeks. There's a million threads on tuning the buffer cache in the forums.

The default buffer cache settings are arguably too generous by default.

[John Bolene]

Don't reboot unless you need to, and the reason given was not a need to.

I only reboot if patches make me, or if the server has to be taken down for hardware or power reasons.

[harry d brown jr]

Jim, I have to agree with Clay's first sentence (I almost fell out of my chair laughing so hard). But, Jim, don't windows admins have to schedule their reboots for at least twice a day, more if the server has any use?

Honestly, I never reboot a server just to reboot it.

Jim, make GREAT use of your Admin's time, and make them update the server once a month with PATCHES (that way they feel like they are useful), and they get to reboot the server (sometimes twice in one install!). If you go this route, make sure they use the custom patch manager process!!! It's a great TOOL!

live free or die
harry

[Darrell Allen]

Hi Jim,

Looks like the concensus is don't reboot unless there is a compelling need to. Just to refresh performance isn't a compelling need unless a degradation in performance is documented. Then you do need to reboot but more importantly, as others have said, you need to find the cause of the degradation.

When I started as a UNIX sysadmin (some years ago), I was told that UNIX was not designed to run for extended periods of time and that UNIX systems should be booted weekly. I'm not sure if there was some validity behind that or not. Perhaps those folks simply had what I've heard called the "mainframe mentality" (must be similar to the "windows mentality").

I do believe there are reasons for regular reboots though I don't see the need for them to be more frequent than quarterly to coincide with installation of current patches.

More comments from my response to an earlier post:

Do you get the feeling there is no generally acceptable rule about how often to reboot? I like to reboot every now and again just to make sure the system will. Yes, systems generally run for long periods of time without needing to be rebooted but changes are made that could keep it from booting. Hopefully you are patching your system on some regular basis. I'd always reboot after installing a patch bundle, even if it is not automatic. I'd rather the system fail to come up in a controlled test than when I'm not testing and I need it back up NOW!.

The above is from:
http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0xfd8b972194d6d5118ff40090279cd0f9,00.html

Darrell

[Darrell Allen]

Hi Jim,

Found some more threads by going to search.hp.com and searching for +reboot +often:

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0xefba6af52b04d5118fef0090279cd0f9,00.html

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0xf679dfe5920fd5118fef0090279cd0f9,00.html

http://forums.itrc.hp.com/cm/QuestionAnswer/1,,0x67f3c6af36b7d5118ff10090279cd0f9,00.html

Darrell

[Santosh Nair_1]

These kind of "solutions" really don't solve any problems. What's to say that your performance problems won't re-appear after rebooting the machine? Its much better to start taking a close look at the machine and seeing what's making the performance bad. Chances are there are some processes that are eating up CPU/memory/io. Remember Glance and top are your friends.

Hope this helps.

-Santosh

[Jim Shepherd_1]

Well, I told our admin that the reboot was not necessary, and here is his reply:

I know that they don't run windows but It is a good practice to reboot the servers weekly/2 weekly/monthly ( Irrespective of operating system ). Also the time between reboots depend on what type of applications they run . If heavy database applications/processes are run then we should reboot weekly or 2 weekly to avoid system hang ups ( as per my experience on (unix)database servers) . Booting undoubtedly improves performance by clearing out cpu registers and data blocks held up by memory ( I am not suggesting booting as a solution for performance gain , there are other methods ). It should not happen that at some point of time the system will hang/stop working when you are running important jobs. Its sys administrator's duty to look after these things but if you have large jobs pending there is no problem in waiting. Since our servers don't run heavy database applications/processes period between two consecutive reboots can be elongated.


I had never heard of CPU registers being locked up by an application. Is this possible?

[A. Clay Stephenson]

Hi again Jim:

After this I definitely would resort to Plan A: get a new admin. - I was joking before but now I'm not. That stuff about CPU registers is total nonsense. CPU registers get clearing/exchanged/modified many,many,many times per second. He might be talking about the TLB (Translations Lookaside Buffer) but really the same thing applies. It too gets 'cleared' many times per second. If the guy doesn't know how to say 'I don't know' but instead resorts to nonsense then he is dangerous. With stable software there is no reason to down a box other than for periodic patches. If he wanted to reboot to make sure the box will reboot then that too is a good enough answer although the answer there is to have enough mirrors so that you don't worry about it. I have literally kept some UNIX servers up for years without a single reboot and somehow they managed to run just fine without 'clearing the CPU registers'.

If a box needs to be rebooted to fix software then the answer is to fix the software not reboot the box.

I truly don't mean to sound harsh but this guy scares me.

Regards, Clay

[James R. Ferguson]

Hi Jim:

Tell your fellow "admin" to wave a dead chicken over the server to banish the bad ether, too.

I think the logic and the empirical data he has provided to support his belief is about as valid.

Regards!

...JRF...

[fg_1]

jim

If this is what your admin is preaching and can back it up, then maybe he has a cure for male pattern baldness, cancer, or can solve the mystery of human cloning while he's at it. NOT asking too much.

On a serious note, he is off the wall with these beliefs, first off unix is a stable o/s not like windows. I think he needs to realize he is not in a windows environment. I would definitely send this individual to the performance & tuning course. Let him go there and spout out some of the things he's saying to you and he will be laughed out of the course.

[Jim Turner]

Hi Jim,

Oh, this one is just too good to let pass by. First off, tell you're "admin" that you'd like to reset *his* CPU registers with a couple hundred volts across his forehead.

I almost coughed up a lung laughing at his nonsense response. Send this guy packing before he causes any damage. Like Clay, I'm not kidding -- I'm dead serious. Hand him a pasteboard box, tell him to pack, and kick him in the arse on the way out.

Oh yeah, also let him know that mauve is the best color for a database. Great performance.

Baaaaaaaaaaaahahahahahaha!!!!!

My system uptimes are measured in years. Your admin is an idiot. Get rid of him.

Cheers,
Jim