Reboot HP to refresh performance?

[Randy Tarrier]

Hi Jim,
I almost died laughing at some of the great responses! Then it struck me where I had heard this before.... I used to manage shipboard Univac boxes in the Navy; these were truly 2nd generation batch-oriented machines with honest-to-good tty's & card-readers! The cpu was 6'x2'x2' and had a whopping 12K RAM!
Due to obvious limitations, our Cobol programs used segmentation to swap code from object lib's stored on tape to RAM. Sometimes, the programs weren't efficient at releasing memory; so we always rebooted after long-running programs to "clean out the registers".
Tell your guy to toss his punched card decks and go buy himself a pc xt, which for him is probably a "great leap forward".

[Bill McNAMARA_1]

Mmm,

There's no harm I suppose, but I'd suggest defragging the disks and adding more memory while you're at it.

Take this as an opportunity to add reboot patches!

Later,
Bill

[Donald Kok]

Is there actualy a way to reboot a unix machine? ;-)

[BFA6]

Jim,

We had a 3rd party supplier (application) who insisted we reboot our servers every two weeks, to "improve performance".

We no longer use this supplier, so don't reboot unless there is a hardware or power problem.

Hilary

[Rita C Workman]

Ah Guys..stop...I'm laughing too hard now !!! I didn't realize this was the humor post of the week...wheww !!

My Dear Mr. Shephard, print this post and give to your new "Admin". Maybe you could also direct him to the www for HP Education. I recommend starting with UNIX Basics and have him stay there and continue on through SysAdmin I & II before he comes back and creates any real damage.
Lord knows he needs education, because if you don't when to reboot...you surely don't belong behind console with a root password doing it !!

Happy trails,
Rita


Mysteries of the Universe...

?? Why is it that every person I know who has created a simple MS Access database NOW thinks they qualify to be an Oracle DBA???
?? Why is it that everyone that found the on switch on a PC thinks they are Computer Technician??
??Why is it that top Management all too often think that anyone who can type can be the new UNIX Admin or..they're MS Certified, so they're qualified to work on UNIX??

I need some coffee....

[John Bolene]

Answer police to action.........

This member has assigned points to 2 of 31 responses to his/her questions.

[Thierry Poels_1]

Hi,

"If heavy database applications/processes are run then we should reboot weekly ...."

LOL. If you reboot your database server you'll loose your database buffer (which can be pretty large on heavy systems). After the reboot the first n transactions will suffer from physical reads for a while, until the database buffer stabilizes again.

regards,
Thierry.

[Kirby Collins]

The short answer is that you shouldn't normally need to reboot an HP-UX system running abaqus to refresh performance.

There are a couple of performance reasons commonly given for rebooting. The first, memory leaks, shouldn't be a problem with abaqus. Abaqus doesn't have a large, permanent shared memory segment (like a database), nor am I aware of any indirect leaks it might cause through HP-UX calls. In any case if you start running short of memory that should be pretty obvious.

A second reason might be to shrink the dynamic buffer cache. With the default settings for the dbc_min_pct and dbc_max_pct kernel tunables, the buffer cache will occupy between 5% and 50% of memory. Unfortunately, the buffer cache has a tendency to grow to the maximum and stay there. This is because the buffer cache algorithms prefer growing the cache over reusing a valid buffer, and pages are not reclaimed from buffer cache until there is memory pressure. Rebooting will not really solve this problem, as the cache will grow right back up to the max after only a few abaqus jobs are run (abaqus scratch files typically run to multiple gigabytes).

For technical servers we normally recommend that the customer fix the size of the buffer cache, either by setting dbc_min_pct and dbc_max_pct to the same value, or using the bufpages tunable (which overrides the dbc parameters). What is the right size really depends on your workload (what you want is to figure out how much memory you need so that all your abaqus jobs avoid paging, then use the rest for buffer cache). If you are interested I have data for abaqus standard and explicit throughput for various buffer cache sizes on a Superdome.

Yet another performance reason given for rebooting is to defragment memory and increase the ability of a program to get large pages. I've never actually run any experiments to examine this, but on the face of it there is probably some truth to it. The buffer cache plays a part here, because it grabs memory in small chunks, and tends not to release it (see above). Fixing the size of the buffer cache at a smaller percentage of memory will avoid too much fragmentation from this source. It is also likely that running user space programs can generate some degree of fragmentation, but I would think this would be a transient effect and a reboot wouldn't help.

Note that page sizes are completely transparent to abaqus...if the OS can't find any pages as large as the hint, then it simply uses smaller ones. The only impact might be a higher TLB miss rate. So I would not in any case recommend rebooting just to increase the availability of large pages.

Kirby Collins
Hewlett-Packard
HPTC Performance Team