RE: 11g RedHat 5 and Hugepages? anybody successful ?
Date: Mon, 11 Oct 2010 19:35:24 -0400
Good advice Andreas- I am going to try your suggestions.
[mailto:oracle-l-bounce_at_freelists.org] On Behalf Of Andreas Piesk Sent: Monday, October 11, 2010 6:25 PM
Subject: Re: 11g RedHat 5 and Hugepages? anybody successful ?
Crisler, Jon schrieb:
> Don, I am way beyond that point. If you set memory_max_target = 0
> start the db, it generates ORA-00849 is thrown.
> ORA-00849 SGA_TARGET value was more than MEMORY_MAX_TARGET value.
> If SGA_TARGET is unset , then I get ORA-27102: out of memory
> Linux-x86_64 Error: 28: No
> left on device
i've no problem with 11g RAC on RHEL5 with hugepages.
just to be sure, you start the instance with memory_target unset,
memory_max_target unset and
i think, the instance does not use hugepages at all and tries to
allocate the memory from the
default 4k pool which fails because of the allocated hugepage pool.
have you tried smaller values like 2G for sga_target (assuming you have
some gigs left after
allocating huge pages)? as others said, creating a pfile for testing saves some time. please check
the memlock ulimit in your shell session before starting the instance. if the instance starts with sqlplus, check from which pool the memory has been allocated
(/proc/meminfo or pmap)
also, there's a bug in CRS. so if you start the instance with srvctl,
your memlock limit will not be
used. the latest PSU was supposed to fix that, but the PSU, at least the CRS part, is broken and
doesn't patch the CRS files at all. in this case you have to patch ohasd manually. Oracle gave me
instructions in the SR i've opened for this issue, i can post them here if the bug could be the
cause of your problem.
if all fails, trace the startup of the instance with strace:
strace -f -e trace=shmget sqlplus / as sysdba
the 3rd argument, shmflg, should be SHM_HUGETLB.