Re: lock_sga error

  • From: Tanel Põder <tanel.poder.003@xxxxxxx>
  • To: "Oracle-L (E-mail)" <oracle-l@xxxxxxxxxxxxx>
  • Date: Sat, 6 Aug 2005 00:23:39 +0100

Re: lock_sga errorHi,

I'm not sure whether Linux supports shared pagetables (which Kevin already 
mentioned), that would help us to keep the kernel virtual<->physical memory 
pagetables smaller and hardware TLBs more efficient and since all processes 
would access locked parts of SGA using the same pagetable entries (as the 
virtual memory pages are known to remain in the same physical memory address), 
we might have less soft page faults, thus improved performance in servers 
having high number of server processes running in them.

But I don't know whether Linux actually supports shared page tables.

But anyway, if you run ulimit -l you should see what's the ulimit for max 
lockable memory for your user, you should set it to at least the size of SGA 
(or parts of the SGA) which you're gonna lock.

Tanel.
  ----- Original Message ----- 
  From: Tim Gorman 
  To: Oracle-L (E-mail) 
  Sent: Friday, August 05, 2005 5:42 PM
  Subject: Re: lock_sga error


  Magnus,

  I was asking why you thought you wanted to get your shared memory to stay 
resident?

  -Tim


  on 8/3/05 1:28 PM, Magnus Andersen at Magnus.Andersen@xxxxxxxxxxxxxxx wrote:


    Thanks Kevin.  I wanted to get my shared memory to stay resident and I did 
implement hugtlb.  I am not using the indirect_data_buffers since I only have 
access to a 1.7 GB SGA at this point.

    Magnus


      -----Original Message-----
      From: Kevin Closson  [mailto:kevinc@xxxxxxxxxxxxx]
      Sent: Wednesday, August 03, 2005 2:54  PM
      To: Oracle-L (E-mail)
      Subject: RE: lock_sga  error


      lock_sga is an OSD init.ora param...it only means  something

      at the port level.. for instance lock_sga was  implemented on the 

      DYNIX/ptx port, but use_ism was  implemented on Sol instead... 

      Either way, the hope is that the OS complements this  param

      with shared page tables and hard-locked SysV  SHMEM...

      neither of which really related to Linux  much.

       
       
      Are you sure you can boot of that init.ora file even  with lock_sga

      commented out?

       
       
      Now, if you want your shared memory to stay resident  (novel

      concept, I know, but it is a bit new to linux) you  have to 

      fiddle around with HUGE_TLB and so on. Also, if using  

      indirect_data_buffers, you will need to use ramfs  over

      shmfs because the latter is pageable  ...

       
       
       

       


         

------------------------------------------------------------------------
        From: oracle-l-bounce@xxxxxxxxxxxxx  
[mailto:oracle-l-bounce@xxxxxxxxxxxxx] On Behalf Of Tim  Gorman
        Sent: Wednesday, August 03, 2005 9:09 AM
        To:  Oracle-L (E-mail)
        Subject: Re: lock_sga error

         
        Why do you want to  do this?



        on 7/25/05 7:43 AM, Magnus Andersen at  Magnus.Andersen@xxxxxxxxxxxxxxx 
wrote:

         

          Hi  All,
           
          I'm trying to lock my sga in memory on a RedHat Linux 3  AS server.  
When I try to start the database with the "lock_sga=true"  paramerter in my 
init file I get the following error:
           
          SQL>  startup nomount;

          ORA-27102: out of memory

          Linux Error: 12:  Cannot allocate memory

          SQL>

          Any  ideas?

          Thanks,

          Magnus Andersen
          Systems Administrator /  Oracle DBA
          Walker & Associates,  Inc.
           







Other related posts: