1836529 Members
3809 Online
110101 Solutions
New Discussion

Re: Ram Avg Util

 
SOLVED
Go to solution
Max ON_1
Valued Contributor

Ram Avg Util

Hello,

I have an HP-UX 11.23 in IA system.
I'm monitoring the performance by MWA agent and SNMP-MIB.
Both the metrics about mem util values are above 80% util. I don't have any critical process that running.
What do you think ? Is it normal ??
Thanks
Max
11 REPLIES 11
A. Clay Stephenson
Acclaimed Contributor

Re: Ram Avg Util

Without knowing how much memory is in your box and without knowing how many (and how large) your "non-critical" processes are, could you answer your own question?

In any event, if your dbc_max_pct was left at the default value of 50, then 50% of your memory will be allocated to buffer cache until competition with processes forces that value down.

If it ain't broke, I can fix that.
Patrick Wallek
Honored Contributor

Re: Ram Avg Util

It depends. What's your definition of normal? If it has been consistently running at that level and has not spiked up suddenly, then it may be "normal" for this machine.

What you don't say is how much RAM you have. 80% of ???? RAM is the real key. 80% of 1GB and 80% of 16GB of RAM are drastically different numbers.

The percentage can also depend on what OS processes are running. Each will take up a percentage of RAM.

Max ON_1
Valued Contributor

Re: Ram Avg Util

Thanks for your reply.
I have 16 GB of physical ram and 6 CPU.
The parameters dbc_max_pct and dbc_min_pct are default value, 50 & 5.
How is possibile, if I don't have any process that is running ??
All application process are stopped and the mem avg util is always same, > 80%.
Is there another kernel parameter for tune the ram configuration ??
Output of top on my system is :
Memory: 2447896K (2238120K) real, 3664904K (3367488K) virtual, 2078656K free Page# 1/6
output of dmesg |grep phys :
physical page size = 4096 bytes, logical page size = 4096 bytes
Physical: 16735472 Kbytes, lockable: 12515580 Kbytes, available: 14534252 Kbytes

thanks in advance
Max
Patrick Wallek
Honored Contributor
Solution

Re: Ram Avg Util

Well, as Clay said, the very first thing I would do is decrease dbc_max_pct. That is taking up approximately 8GB out of your 16GB of RAM. I would set it down to a value of 8 or 10. That will help you immensely.

There isn't really any other kernel parm to "tune" your RAM in the manner you seek.

The only other thing to do is to go through the processes on your system and see what is using RAM. The easiest tool to do this with is glance/gpm.

Without seeing a list of running processes and their size, it's very difficult to give any advice.
A. Clay Stephenson
Acclaimed Contributor

Re: Ram Avg Util

First, top is absolutely useless as a memory tool because it knows nothing about non-process memory such as buffer cache. Stick with MeasureWare/Glance as your diagnostic tool. With your dbc_max_pct at 50, that's 50% of your 80%. Although you have stopped all your applications there are still many UNIX processes running so that's a bit more. What I would really look for are things like shared memory segments. Even though no applications are running, you may still be using large amounts of shared memory either by application design or because you improperly terminated an application before it could clean up. Can you say kill -9?

Bear in mind that there is nothing wrong with nearly 100% memory utilization as long as you see no page outs. 11.23 does well with large buffer caches but you will get things to appear better if you tune down the buffer cache to 20% or so.
If it ain't broke, I can fix that.
Steven E. Protter
Exalted Contributor

Re: Ram Avg Util

Shalom Max,

Think about this. There is nothing intrinsicly wrong with using memory.

I mean it was purchased,money was paid for it. So it should be idle and not in use?

Is the default buffer cache setting right for your system? Probably not. I usually dial down buffer cache by dropping dbc_max_pct to 7.

Leaves memory for other things. Like Oracle, which I work with extensively.

I sense that you are measuring things but perhaps for no reason. Bill Hassell (and I a good parrot) advise not to start trying to solve performance problems, until you you actually have evidence of a performance problem.

Do you have a problem?

Some cool perf measurement tools:

http://www.hpux.ws/system.perf.sh

SEP
Steven E Protter
Owner of ISN Corporation
http://isnamerica.com
http://hpuxconsulting.com
Sponsor: http://hpux.ws
Twitter: http://twitter.com/hpuxlinux
Founder http://newdatacloud.com
Max ON_1
Valued Contributor

Re: Ram Avg Util

Thanks to all for explanations.
I could change the value of dbc_max_pct for to see the util mem percent more down.
But I have OVPI with Oracle on my system and I don't understand if this change involve less resources for my application.
James R. Ferguson
Acclaimed Contributor

Re: Ram Avg Util

Hi Max:

> I could change the value of dbc_max_pct for to see the util mem percent more down.
But I have OVPI with Oracle on my system and I don't understand if this change involve less resources for my application.

Oracle does its own buffering and thus allowing the standard Unix file buffer cache to consume 50% of your memory is just doing double-buffering. You should work with your DBA (or vendor) and insure that Oracle has the buffers it needs. Tune down, as suggested, your Unix buffer cache as suggested.

Regards!

...JRF...
Steven E. Protter
Exalted Contributor

Re: Ram Avg Util

Shalom,

The reason dbc_max_pct is important is that with Oracle on 11.11 double buffering occurs.

This provided no performance benefit and when the Oracle SGA was increased perofmance improved.

Bill Hassell recently reported to me via email that buffer cache being higher under some circumstances did provide a performance boost to Oracle running on cooked filesystems.

This is one component of overall performance on a system.

SEP
Steven E Protter
Owner of ISN Corporation
http://isnamerica.com
http://hpuxconsulting.com
Sponsor: http://hpux.ws
Twitter: http://twitter.com/hpuxlinux
Founder http://newdatacloud.com
Bill Hassell
Honored Contributor

Re: Ram Avg Util

The default value of 50% for the buffer cache is simply a bad value. Since you are running 11.23, the buffer cache code was heavily updated to improve performance so that the recommended maximum for dbc_max_pct could be larger than 800-1500 megs. For a very busy system (at 11.23), 4GB to 6Gb can be useful.

But your metric 80% isn't meaningful. You need to define the amount of local process memory, shared memory and the buffer cache. 80% is not right if you paid lots of money for your RAM...a better value is 95% or higher. After all, this is not a PeeCee where the system becomes unstable or crashes when all memory is used. If you have terminated all your applications, then local process memory will be very small but if you trashed the applications by using kill -9, then shared memory and other IPCS objects are left orphaned in RAM.


Bill Hassell, sysadmin
Max ON_1
Valued Contributor

Re: Ram Avg Util

OK Bill
thank you very much. I'm tuning my system and I changed the value of dbc_max_pct. I'm monitoring the system resource and my applications and it's seems that works very well.

thanks again to all

regards
Max