How to release unused "Message Buffers"?

[Rodrigo Aoki]

I have a rx2620 running HP-UX 11.23 and two SAP instances and I want/need to redistribute some (unused) system memory.

I observed with Glance (System Table session) that there are 1.5 GB allocated to "Message Buffers" and 0 KB used.

System Table Available Requested Used High
--------------------------------------------------------------------------------
Inode Cache (ninode) 214560 na 2394 2394
Shared Memory 0mb 65.4mb
Message Buffers 1.50gb na 0kb 0kb

Does someone know how I can release this unused memory?

I didn´t find which kernel parameter are related to this buffers. The "kctune | grep msg " returned the above configuration.

msgmap 2048 2048
msgmax 8192 Default Immed
msgmnb 65535 65535 Immed
msgmni 512 Default
msgseg 32767 32767
msgssz 96 Default
msgtql 2046 2046

[whiteknight]

Rodrigo,

check your dbc_max_pct ? Default value is 50, if you reduce this, you will get more memory

WK

[Armin Kunaschik]

Who did tell you, that this is unused memory??
SAP and Oracle need (much!) IPC resources to work.

Use ipcs -a to determine what IPC resources are used (look for "m" in the first column)!
If your SAP is stopped and you still see resources owned by SAP users, you need to manually remove them using ipcrm.

If SAP is still running there is no easy way to find orphan resources...

If you don't find anything with this procedure and the system starts paging (swapinfo -tam, vmstat), consider buying more memory!

My 2 cents,
Armin

[Rodrigo Aoki]

WK,

The dbc_max_pct parameter is set to 10 percent. Since the machine has 6 GB of physical memory maybe I "lost" 600 MB to filesystem page buffers.

This is not the same value of Message Buffers. Could you explain me what are the Message Buffers?

[Rodrigo Aoki]

WK,

The dbc_max_pct parameter is set to 10 percent. Since the machine has 6 GB of physical memory maybe I "lost" 600 MB to filesystem page buffers.

This is not the same value of Message Buffers. Could you explain to me what are the Message Buffers?

[Rodrigo Aoki]

WK,

The dbc_max_pct parameter is set to 10 percent. Since the machine has 6 GB of physical memory maybe I "lost" 600 MB to filesystem page buffers.

This is not the same value of Message Buffers. Could you explain to me what the Message Buffers are?

[Frank de Vries]

use ipcs -q
for message queue and id

and ipcrm -q
to remove messqge queue
You may need to wirte a for loop if you have a lot of them.

regards,

[Rodrigo Aoki]

Sorry for the repeated replies. The HP´s site returned to me unsuccessful postings.

I checked the message queue of the server and compared with another servers with different capacities (the physical memory varying from 6 GB, 8 GB, 11 GB, 40 GB) and all the machines were configured (?) with 1.5 GB for the Message Buffers.

So the 1.5 GB value is not related to a parameter which uses a percent size of the memory.

Also I find a server with 10.8 GB of Message Buffers, 3.6 GB of free memory and 8 GB of physical memory. The value for QBYTES of the message queues (ipcs -qa) are the same than the other group of machines.

So message queue buffers are not the same than Message Buffers.

But it´s very strange the value of 10.8 GB since there are only 4.4 GB allocated...

Now I looking for an better explanation of what is the "Message Buffers" and maybe for a knowed bug of measurings got by Glance...

Also I´ll check another machines with different architectures (RISC, not Itanium/EPIC) to comparing the values...

Does anybody have another idea?

[Rodrigo Aoki]

Hi Folks,

Humm... I think that I should publish my apologies.

The values (showed by Glance) for a Itanium production machine are fabulous... if they REAL... "Available" is the TOP VIRTUAL limit of the memory segments and NOT the actually "Used" or "Requested" values.

System Table Available Requested Used High
--------------------------------------------------------------------------------
Inode Cache (ninode) 8192 na 2435 2536
Shared Memory 6.25tb 1.97gb
Message Buffers 1.50gb na 0kb 0kb
Buffer Cache 3.20gb na 3.20gb na
Buffer Cache Min 2.00gb
Buffer Cache Max 3.20gb
DNLC Cache 8976

Model : ia64 hp server rx864 Phys Memory : 40.0gb Network Interfaces : 6
OS Name : HP-UX Number CPUs : 10 Number Swap Areas : 4
OS Release: B.11.23 Number Disks: 218 Avail Volume Groups: 4
OS Kernel Type: 64 bits Mem Region Max Page Size: 4.00gb

Once I have seen the "Available" 1.5 GB, this value was NOT allocated... and I haven´t unused system memory to redistribute...

[Rodrigo Aoki]

Finally, the Message Buffer LIMIT of 1.5 GB is equal to the group of machines that I checked because ALL the machines are serving SAP instances.

So, they use some recommended value for SAP.

And a tip: despite the ipcrm, any SAP instance contains a program called "cleanipc" which can clean all the ipc segments not released from a SAP instance AFTER its normal shutdown. When you have done a "quick" restart (without any changes of SAP parameters) and the instance could not start, try to release the shared memory with:

cleanipc

Example: cleanipc 05
For the SAP instance that uses system number 05, or TCP ports 3205, 3305 and 3605.

[Frank de Vries]

Hi Rodrigo

Any points for our efforts ?