Re: cache buffer chains/where in code

Ok, next tip:

In my experience queries attacking the same buffer(s) show up in ASH and AWR reports as being executed lot of times and being quite heavy on buffer gets.
Especially watch out for the *same query* executed lot of times at the *same time* by *many sessions*.
Once I just proofed a simple case:

for a given query taking hundreds buffers gets per row and returning about 15 rows
and for a given hardware
it took about 30 parallel sessions to explode "cache buffer chains" waits. Up to 30 sessions the server performed ok. The same query managed to perform about 60 sessions in parallel on a better server.

This is the point: scalability works only this far for a particular load on that particular hardware.

The other point is to look into application design and eliminate many database sessions doing essentially the same task again and again. I my case it was many user sessions polling the same data at some 2-3 seconds interval just to get "real time" data display on their screens. One day number of users increased to the point where server just stood still.

The solution was both to reconsider polling interval and the design in general and to tune the query.

Please consider the environment before printing this e-mail

                                                                           
             Christo Kutrovsky                                             
             <kutrovsky.oracle                                             
             _at_gmail.com>                                                To 
                                       Laimutis.Nedzinskas_at_seb.lt          
             2009.12.11 17:20                                           cc 
                                       oracle-l <oracle-l_at_freelists.org>   
                                                                   Subject 
                                       Re: cache buffer chains/where in    
                                       code                                
                                                                           
                                                                           
                                                                           
                                                                           
                                                                           
                                                                           

There are no connects/disconnects at that time.

On Fri, Dec 11, 2009 at 2:16 AM, <Laimutis.Nedzinskas_at_seb.lt> wrote:.

      hehe, I concur.
      And by the same chance are there many connects and specially
      disconnects
      going on in a short time?

      ---------------------------------------------------------------------------------


      Please consider the environment before printing this e-mail





                  Greg Rahn

                  <greg_at_structuredd

                  ata.org>

      To


                  Sent by:                  Christo Kutrovsky

                  oracle-l-bounce_at_f         <kutrovsky.oracle_at_gmail.com>

                  reelists.org

      cc


                                            Martin Berger

                                            <martin.a.berger_at_gmail.com>,

      Tanel


                  2009.12.11 01:24          Poder <tanel_at_poderc.com>,

      oracle-l


                                            <oracle-l_at_freelists.org>


      Subject


                  Please respond to         Re: cache buffer chains/where

      in


                  greg_at_structuredda         code

                       ta.org










      By chance is this using UFS file system and not using directio
      (forcedirectio)?

      On Thu, Dec 10, 2009 at 11:45 AM, Christo Kutrovsky
      <kutrovsky.oracle_at_gmail.com> wrote:
      > I traced down the problem to a Solaris 10 BUG. bug_id 6642475, has
      to do
      > with kernel locks when trying to allocate contiguous memory. The
      code is
      > inefficient, and the workaround is to disable it via echo
      > "pg_contig_disable/W 1" | mdb -kw.
      >
      > I hope this helps someone out there. I don't know in what release
      it is
      > resolved.

      --
      Regards,
      Greg Rahn
      http://structureddata.org.
      --
      http://www.freelists.org/webpage/oracle-l




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

--
Christo Kutrovsky
Senior Consultant
Pythian.com
I blog at http://www.pythian.com/blogs/

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