Experience with Solaris 10 Memory Management

[Former Member]

Dear Forum!

Using Note 724713 we installed a couple of R3 Systems on Solaris 10. Independently which R3 Version is used we observe much more memory problems as with prior solaris releases or other operating systems like AIX.

For instance1: We moved three R3 Systems from a server with solaris 8 to a server with solaris 10. We didnt change memory parameters of R3 and the databases (oracle). The solaris 10 server has the same amount of memory (8GB) like the solaris 8 server. Now, on the solaris 10 server we can only start two R3 systems, the third one doesnt start because no memory can be allocated. The memory parameter of all component slightly overload the main memory. Because the systems are not productive they mostly idle and performance loss due to paging and even swapping didnt worry us. Not even that the third system didnt start, the whole OS was a kind of freezed and no more work was possible, also not on the other two R3 systems which were already startet.

For instance2:

On another server with 16GB memory we startet an installation of a sapsystem and the sapinst crashed due to memory absence. Admittedly, here the main memory was also a bit overloaded, but this is not unusal here and I never saw sapinst crashing because of this.

In both cases Swapspace was adequate.

Do you have similar experiences? Or any tipps to solve the problem.

Thanks

Andreas

[Former Member · Accepted solution]

Hi,

Have you referred to the note 300464 especially the line about the project.max-shm-memory from the note.

Regards,

Rohan.

[markus_doehr2]

If you just copied the parameter from /etc/system of the old server to the new one they won't have any effect.

As already mentioned you need to configure a project file (/etc/project) and assign the users to the respective projects in /etc/user_attr.

We run many instances on Solaris hosts (up to 12 on one system) using zones, great concept and very administrator friendly because all the zones appear as separate "machines" rather than separated with instance numbers. Using resource management you can even assign a number of CPUs to a certain zone/instance.

Make sure you add

set pg_contig_disable=1

to /etc/system to avoid the system copying huge amounts of memories around to free enough contigous memory to start a new system.

Markus

[Former Member]

Hi, we set the memory parameters like in note 724713

projadd -p <proj_id> -c "SAP System <SID>" \

-U <sid>adm,<db_sid>,sapadm \

-K "process.max-file-descriptor=(basic,65536,deny),(priv,65536,deny)" \

-K "process.max-sem-nsems=(priv,2048,deny)" \

-K "project.max-sem-ids=(priv,1024,deny)" \

-K "project.max-shm-ids=(priv,256,deny)" \

-K "project.max-shm-memory=(priv,18446744073709551615,deny)" \

<proj>

Is this good enough?

What we didnt do ist set pg_contig_disable=1

I try out if this helps and let you know

Thanks so far

Andreas

[Former Member]

this is strange. Although /etc/user_attr shows

/etc/user_attr shows

#

adm::::profiles=Log Management

lp::::profiles=Printer Management

postgres::::type=role;profiles=Postgres Administration,All

root::::auths=solaris.*,solaris.grant;profiles=Web Console Management,All;lock_after_retries=no;min_label=admin_low;cle

arance=admin_high

qp7adm::::project=QP7

oraqp7::::project=QP7

sapadm::::project=QP7

smdadm::::project=QP7

the process list show that the processes of the user sidadm run in project "system" only the oracle user project is ok:

root@pi7q # ps -eo user,pid,ppid,project,args

USER PID PPID PROJECT COMMAND

oraqp7 1783 3652 QP7 ora_q001_QP7

oraqp7 125 3652 QP7 ora_mmon_QP7

qp7adm 1884 1862 system dw.sapQP7_DVEBMGS00 pf=/usr/sap/QP7/SYS/profile/QP7_DVEBMGS00_pi7q

oraqp7 20464 3652 QP7 oracleQP7 (LOCAL=NO)

daemon 4416 3652 system /usr/lib/nfs/nfs4cbd

qp7adm 1870 1862 system icman -attach pf=/usr/sap/QP7/SYS/profile/QP7_DVEBMGS00_pi7q

oraqp7 20462 3652 QP7 oracleQP7 (LOCAL=NO)

root 14658 25510 user.root bash

oraqp7 105 3652 QP7 ora_dbw1_QP7

oraqp7 2980 3652 QP7 oracleQP7 (LOCAL=NO)

root 5819 3652 system ./tGAgent.d -d

oraqp7 2544 3652 QP7 oracleQP7 (LOCAL=NO)

smdadm 5639 3652 system /usr/sap/SMD/SMDA02/exe/sapstartsrv pf=/usr/sap/SMD/SYS/profile/SMD_SMDA02_pi7q

qp7adm 1900 1862 system dw.sapQP7_DVEBMGS00 pf=/usr/sap/QP7/SYS/profile/QP7_DVEBMGS00_pi7q

root 4443 3668 system /usr/lib/saf/sac -t 300

.

.

.

.

.

[markus_doehr2]

If you post such outputs, please use the code tag (the << >> button next to underline) around them so they are readable here on the web.

> the process list show that the processes of the user sidadm run in project "system" only the oracle user project is ok:

Most likely the problem is, that after the creation of the /etc/project files the parent process "sapstartsrv" was not restarted; since the SAP system uses the environmental settings of that process (environment is inherited) the system doesn't use the actual project.

Try

- stop the system

- stop saposcol (saposcol -k)

- execute

/etc/init.d/sapinit stop

check if all processes are stopped.

Start the system again.

Markus

[Former Member]

thanks, this was it