necessary kernel changes after memory upgrade???

[derek b smith_1]

we went from 256mb to 2gb of memory on a DNS server. Is there any needed kernel changes such as maxdsz or any other parameter?

thank you
derek

[Ninad_1]

Derek,

As you have increased memory
1. You need to increase your swap space
2. You may reduce the dbc_max_pct kernel parameter - if set to default 50% - can reduce to say 10%-15% depending on how much buffer cache you need - which equals to 200MB-300MB out of your 2GB memory
3. You may increase the maxdsiz, maxssiz, maxtsiz and the 64 bit params - maxdsiz_64 ... only if you need additional data/stack/text space for your programs otherwise not necessary.

Regards,
Ninad

[A. Clay Stephenson]

Because HP-UX is a virtual memory machine the amount of physical memory in the box really has nothing to do with the process size limits. If you are running dynamic memory cache then you should perhaps looks at max_dbc_pct and min_dbc_pct because almost a 10X increase in physical memory will change those values very significantly. You may also need to increase maxswapchunks and add more swap space or if your 2GB physical memory eliminates swapping altogether then you may wish to enable pseudoswap (swapmem_on=1).

[derek b smith_1]

ok guys... thanks and this is what I thought. Although I did run a chatr on /usr/sbin/named and it shows as a shared executable. So with that in mind is it still true to say that this process will only ever get up to ~940MB unless it was linked as a EXEC_MAGIC which then it could get upto 1.9gb?

dbsmith@DUBHADM1# chatr /usr/sbin/named
/usr/sbin/named:
shared executable
shared library dynamic path search:
SHLIB_PATH disabled second
embedded path enabled first /usr/lib:
shared library list:
dynamic /usr/lib/libdld.2
dynamic /usr/lib/libc.2
dynamic /usr/lib/libnsl.1
shared library binding:
deferred
global hash table disabled
plabel caching disabled
global hash array size:1103

[Ninad_1]

Derek,

You are right. If your named is 32-bit then you have the limitation of addressing 1st quadrant of memory.
If you see towards the end of chatr named [which you have not posted] will show you if the third quadrant private data space and fourth quadrant space is enabled for use for named - most certainly would not be enabled.
But why do you need the named to access more data area - are you getting any errors ? or how much data area can you see the named consuming from glance and then doing a s , then giving the pid of named and then pressing M.
You can also have a look thru this thread
http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=983809

Regards,
Ninad

[Kent Ostby]

Derek --

To be specific, HP recommends that you have 2x RAM worth of device swap so you need 4 GB of RAM worth of device swap or you may run into problems.

Best regards,

Oz

[Patrick Wallek]

Oh Baloney!!!!!!!!!!! The 2xRAM for swap space recommendation is years out of date!

I don't have ANY machines with that much swap space. Eating up that much disk for what should never be used is ridiculous.

I set up primary swap to be about 25% of my total RAM Size and then set swapmem_on to 1. From there I monitor usage. I also set the maxswapchunks kernel parameter to its maximum value, just in case I need to add more space in the future.

I honestly can't remember the last time any of the machines actually paged out to any significant degree.

[Steven E. Protter]

Shalom Derek,

Just make sure swap is greater than half RAM to handle a dump and call in the morning.

I'd take it up to 1 times ram and see if any actual paging happens.

SEP

[Patrick Wallek]

For further reading on managing swap space, have a look at the following link:

http://docs.hp.com/en/B2355-90950/ch06s03.html

That is the "Managing Swap and Dump" section from "Chapter 6. Administering a System: Managing Disks and Files" of the "Managing Systems and Workgroups" manual.

The main manual is available here:
http://docs.hp.com/en/B2355-90950/index.html

The key sentence is: "Typically, unless the amount of physical memory on your system is extremely large, the minimum amount of swap space should equal the amount of physical memory on the system. "

Now the next sentence does say: "In general, make swap space to be roughly two to four times your physical memory."

I disagree with that. With current machines and a decent amount of RAM, one should not need that much swap space since one should not be paging out.

[derek b smith_1]

thank you for all the replies!
The question remains is named considered a user process or a system process and therefore is it releveant to modify maxd_siz? I say no but a co-worker says yes.
Is there a metric within glance I can pull via gpm script that will tell me if a process is using maxd_siz, maxs_siz or mazt_siz? Or any other tool to tell me this?

thanks again
derek

[Ninad_1 · Accepted solution]

Hi Derek,

I have mentioned how to see how much data,stack and text is being used in my earlier reply.

do a ps -eaf | grep named and note the pid
then
glance
Press s and enter the pid no
Press M [ Note capital M]
This will show you the data, stack, text etc.

As far as named is concerned - my opinion is that first see whether you require to increase the limit for data segment in the 1st place - thru glance.
Next thing can you increase the limit - I think you can do, but any upgrades will overwrite this.

Regards,
Ninad

[derek b smith_1]

question answered.