Re: excessive swapping during rman backups on Linux 2.6

On Mon, 15 Jan 2007 10:48:05 -0800, DA Morgan <damorgan_at_psoug.org> wrote:

>> sysctl:
>> fs.file-max = 65536
>> kernel.shmall = 1572864
>> kernel.sem = 1000 32000 100 128
>> kernel.shmmax = 6442450944
>> net.ipv4.ip_local_port_range = 1024 65000
>> vm.nr_hugepages=3100
>>
>> .......
>> We run Oracle 9iR2,10gR1/2 on RH4/RH3 and Solaris 10 (Sparc)
>> remove NSPAM to email
>
>Run the following as root
>
>/sbin/sysctl -p
>
>and compare with Oracle's recommendations which will be something
>close to this:
>
>fs.file-max = 65536
>kernel.shmall = 2097152
>kernel.shmmax = 2147483648
>kernel.shmmni = 4096
>kernel.sem = 250 32000 100 128
>net.core.rmem_default = 1048576
>net.core.rmem_max = 1048576
>net.core.wmem_default = 262144
>net.core.wmem_max = 262144
>net.ipv4.ip_forward = 0
>net.ipv4.conf.default.rp_filter = 1
>net.ipv4.ip_local_port_range = 1024 65000

Daniel, all the rmem/wmem parameter were actually recommended for 10g dbs.. I fail to see how any of these will help with my IO problem though :)

http://www.puschitz.com/TuningLinuxForOracle.shtml#SettingSharedMemory

(how the first 4 are set on linux:
kernel.sem = SEMMSL SEMMNS SEMOPM SEMMNI Here's the recommendations for 9i on linux x64 (metalink note) SEMMSL 250 (processes + 10, ours is 1000) SEMMNS 32000 (processes across all instances. total semaphores) SEMMNI 128 (max # of shared memory segments, we have this set ot recommended)
SEMOPM 100 (max # of of semaphore ops, we should've set tihs to 250 per link above)
SEMVMX 32767 (semvmx specifies the maximum possible semaphore value, kernel default)
SHMMAX 2147483648 (max sga, our sga is 3G, or shmax is 6g) One-half the size of your system's physical memory. Check your system for additional restrictions.
SHMMNI 4096 (This parameter sets the system wide maximum number of shared memory segments)
SHMALL = SHMMAX/pagesize (our matches
FILE-MAX 65536
IP_LOCAL_PORT_RANGE 1024 65000 The only thing we have probably set wrong is this: "Free system memory will automatically be decreased by the size of the Huge Pages pool allocation regardless whether the pool is being used by an application like Oracle DB or not: "

However, I tried decreasing those during RMAN level 0, and it didnt seem to help

.......
We run Oracle 9iR2,10gR1/2 on RH4/RH3 and Solaris 10 (Sparc) remove NSPAM to email Received on Mon Jan 15 2007 - 19:28:26 CST