Re: 11g RedHat 5 and Hugepages

  • From: Kevin Closson <ora_kclosson@xxxxxxxxx>
  • To: Jon.Crisler@xxxxxxx, tgascard@xxxxxxxxx
  • Date: Tue, 12 Oct 2010 12:19:38 -0700 (PDT)

Don't bother with ASM. Not needed and best to leave it to AMM so it won't be 
able to allocate hugepages anyway. Just focus on your database instance.

The following is where google should have taken you: 
http://kevinclosson.wordpress.com/2010/03/18/you-buy-a-numa-system-oracle-says-disable-numa-what-gives-part-i/


In your case it would be good to read the whole post. That answer to your 
problem (based on my reading of this thread)  is under the "What Went Wrong" 
section.

Let me know what you find out.





________________________________
From: "Crisler, Jon" <Jon.Crisler@xxxxxxx>
To: Kevin Closson <ora_kclosson@xxxxxxxxx>; tgascard@xxxxxxxxx
Cc: Oracle-L@xxxxxxxxxxxxx
Sent: Tue, October 12, 2010 11:33:45 AM
Subject: RE: 11g RedHat 5 and Hugepages

 
Actually Kevin- I ran across your updated blog posts on hugepages last night 
and 
was reading them twice JSo far I have not found any silver bullet.  One of my 
coworkers was speculating that it might have something to do with ASM- we are 
running a mixed 11.2 ASM/CRS with 11.1 db config (due to compatibility issues 
with the app).  My opinion is that it would not matter- either ASM is in huge 
page memory, or its not- as long as there is memory to support it I think it 
would not matter.  The libnuma symlinks are new to me, I was unaware of any 
issues.
 
From:Kevin Closson [mailto:ora_kclosson@xxxxxxxxx] 
Sent: Tuesday, October 12, 2010 2:27 PM
To: Crisler, Jon; tgascard@xxxxxxxxx
Cc: Oracle-L@xxxxxxxxxxxxx
Subject: Re: 11g RedHat 5 and Hugepages
 
hugetlb_shm_group is not needed on Red Hat

Have you googled to see whether I've already blogged the solution to you 
problem?

If you are on 11.1 you probably don't have the libnuma symlinks right. If you 
can't find it with google let me know

Also, this notion that 32bit OS deployments don't need hugepages is quite 
wrong. 
Any system with large connection count is not going to be able to afford the 
wasted page table memory.

Use hugepages.
 

________________________________
 
From:"Crisler, Jon" <Jon.Crisler@xxxxxxx>
To: tgascard@xxxxxxxxx
Cc: Oracle-L@xxxxxxxxxxxxx
Sent: Tue, October 12, 2010 7:45:03 AM
Subject: RE: 11g RedHat 5 and Hugepages

This is pretty much what I am doing, however the parameter-
vm.hugetlb_shm_group=1300
- I am under the impression that this does not apply to RedHat 5, only
to SUSE and a few other varients.
Can anybody that has done this on RedHat 5 verify if this parameter is
needed in /etc/sysctl.conf ?

-----Original Message-----
From: tgascard@xxxxxxxxx [mailto:tgascard@xxxxxxxxx] 
Sent: Tuesday, October 12, 2010 3:48 AM
To: Crisler, Jon
Cc: Oracle-L@xxxxxxxxxxxxx
Subject: 11g RedHat 5 and Hugepages


  Hi,

I did it and is working perfectly well.

My configuration :

Not Using AMM with ASM and database

/etc/sysctl.conf :
vm.nr_hugepages= 12288 -- page size 2MB => 24 GB
vm.hugetlb_shm_group=1300 --id dba, be careful if it's a different user 
between ASM and Database
kernel.shmmax=25769803776 -- => 24 GB
kernel.shmall=6291456 -- page size 4KB =>24 GB

/etc/security/limits.conf :
memlock=25165824 -- KB => 24 GB

/etc/profile
ulimit -u 16384 -n 65536 -l 25165824

To check you are using Huge Page

grep Huge /proc/meminfo

I hope it 'will help.

Thierry.





--
//www.freelists.org/webpage/oracle-l


      

Other related posts: