Yes I know the times I picked are now not matching reality any more. However I think I do mention in the presentation (maybe i’ve forgoten this time) that the exact times are not as important as the ratio of difference between the different memory types.

Stats:
AIX 5.3, P590 series. 8 CPUs, 27GB RAM allocated for the machine partition.
DB SGA 4GB
MAXCLIENT=MAXPERM% = 20% (MIN IS 10%)
lru_file_repage =1
strict value for maxperm and for maxclient both set to 1.
File system=jfs2.
paging = 1% all the time (means no paging)

Consider the following at point in time:
From vmstat -v, it is clear the file cache is at 19.8% meaning almost all of 20% of 27gb = 5.4gb is in used by file cache for jfs2.

From svmon and vmstat , free memory is 6gb.
From nmon also it says total mem=27gb, free=6gb and used=21gb and file cache used is 5.4 gb.

I would like to know where i am doing wrong below:
Objective:
Try to add the os processes memory usage + shared segments +freemem,
and sum should map to 27gb (which is the total memory).

Tot_memory = File_cache + unixprocesses + shared segment memory +freemem
27gb = 5.4 gb + 9gb + 4gb +6g (as you see i am missing 2.4gb uncounted for)

Running the above at different times more or less am short of 2-2.4 gb always. Where is this 2.4 gb and who is using it? If you say to me that this 2.4 gb is actually persistent memory from disk files and the freemem shown my vmstat is actually not complete true, then i need to find out how to calculate this persistent memory on my machine and that better sum to 2.4 gb so that i can reconcile the total mem.

Filecache, freemem and shared segment calculation is pretty much straight forward, and i think you all know how it is to be calculated. (I even checked ipcs -b and i do not have a memory segment orphan etc).

How I come up with 9gb is simple.(9gb is sum of all unix processes memory usage)
Running ps vg command and adding (rss-trs)/1024 (mb) to sum for those process ids where PID is also available in (select pid from v$process).
To this i add sum of rss for rest of the remaining process(oracle listener, dbconsole, and root processes etc).

Oracle processes adds up to major like 8gb, rest 1gb is for all remaining processes.

What about jave heap, is this a separate memory that wont come in ps vg and also in ipcs -b?? Am at a loss.

My only obj for now is to find ways to add all process memory + shared segments + free + file cache and this should sum to 27gb or at least the difference should not be so high as 2.4 gb missing.
I checked i don’t have a leaking unless u tell me how to check??

Cheers
shahid

CAN YOU PLEASE SEND THE REPLY TO MSHAHID@ALFRANSI.COM.SA OR ORACLEFUSIONS@GMAIL.COM

You can see that you have allocated 8193 hugepages to something, so on your system, unless something else is using 10g of ram, the chances are it’s oracle.

Christo- as far as I can see it matters not to set the hugetlb_shm_group - oracle will be able to grab hugepages, certainly i see this behaviour on Redhat 4 Update 3 2.6.9-34.ELsmp kernel. (64-bit).

HugePages_Total: 3250
HugePages_Free: 114
Hugepagesize: 2048 kB

cat /proc/sys/vm/hugetlb_shm_group
0

my /etc/security/limits.comf is also unedited, and I’m still allocating - only when using srvctl though.

Except my only allocate via srvctl also only works when setting both the group and the security limits. This is RAC though.

Looks like it’s used since you have less free pages then taken. Which means *something* has taken them.

You can do the math to be sure:

do a “ipcs” then look how big the shared memory segment is, and compare it to the memory taken (in your case, 10240 - 2047 = 8193*2 = 16386 mb).

SUSE users: beware, suse does “lasy” allocation of hugepages. Which can be a problem if you start multiple databases that wont fit if they touch all it’s SGA.