Memory High on N4000

[Ade Odimayo_1]

Hi Admins,

We noticed sudden increase in memory usage (92 percent) on our N4000 server 4GB RAM with 4 CPUs. We are running four oracle instances requiring approx 2.0GB memory allocation. There are very few users on the databases.

I have attached docs with current kernel parameter, sar and ipcm output
(inod-sz is showing zero is this ok?)

Or do we just have to buy more RAM?

TIA

Ade.

[Paul Hawkins · Accepted solution]

Check the memory usage of the mib2agent (if you have one). There is a known problem with this (memory leak) which can be fixed by applying patch PHSS_26138.

[Stefan Farrelly]

Theres not much to go on in your attachment. How about some output from;

swapinfo -mt

and

UNIX95= ps -e -o vsz=Kbytes -o ruser -o pid,args=Command-Line | sort -rnk1

The second command lists processes in order by how much memory they are using (approx) - largest users first.

[Pete Randall]

You say this was a sudden increase. What changed? Did you apply patches? Start another instance? Add an application? Implement some new feature? Change hardware in some way? Change configuration or kernel parameters?

Memory usage shouldn't just suddenly change with no reason. If you changed something that caused the increase and can back off the change then you're all set. If whatever the change was is required then you're going to need more memory to support it.

Pete

[Ade Odimayo_1]

We increased the nflock value on the kernel parameter.

see attached docs for swapinfo and UNIX95 output

TIA

[Bill Hassell]

Can't tell too much from the ps listing using UNIX95 since the fields are not identified. However, 4Gb is a minimum for 11.0 64bit and the actual requirements for each Oracle instance are defined by the DBA, most specifically, the size of SGA for each instance. A simple change to the INI file for Oracle could increase memory usage 500%.

NOTE: using 100% of memory is a good thing, not a bad thing. When things get slow due to paging (stopping programs and rolling them out to swap), that's bad for performance, so look at the page-out rates in vmstat. If they are 0 or any single digit, you're fine.

With 4 instances, you may want to plan for the future and look at 6-8 Gb when you have more users and the DBAs find even faster ways to access the database by using lots of shared memory.

[Stefan Farrelly]

Your Oracle databases themselves are using just under 2Gb of memory for themselves to serve user requests. However, you have more than just a 'few' users connected to the databases and each one is using a fairly large amount of memory (the LOCAL=NO proceses from the Unix95 command). This first field shows the amount of memory they are using which for your users is around 15-30MB EACH. You soon run out of memory with users using this amount of memory each. I count just over 100 users at around 20MB each = 2 GB ! which when added to the 2Gb Oracle is using soon = your almost 4Gb used.

[Ade Odimayo_1]

Stefan,

What tool did you use to analyse the LOCAL=NO file size by user?

Does anyone know any kernel parameter I can tune to reduce the file size that can be generated per user?

Many thanks.

TIA

[Pete Randall]

You could always return nflocks to it's original value. Why did you increase it, anyway?

Pete

[Ade Odimayo_1]

We have decided to buy extra 4GB RAM.

Could anyone shed any light whether it is normal to have zero inod-sz activity as shown from the sar output attached.