Oracle FAQ Your Portal to the Oracle Knowledge Grid
HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US
 

Home -> Community -> Usenet -> c.d.o.server -> Re: lots of waiting on 'db file parallel write'

Re: lots of waiting on 'db file parallel write'

From: peter <p_msantos_at_yahoo.com>
Date: 28 Feb 2006 06:46:05 -0800
Message-ID: <1141137965.175172.252750@j33g2000cwa.googlegroups.com>


Jon,
  thanks for the feedback.
  We do have on average 10+ concurrrent processes that are doing lots of IFFS and Tablescans...these
  are simply large selects and hash joins...I also don't think this is the problem because these processes
  are running faster than ever..so nobody is complaining... Although we are growing and more and
  more customers are executing these types of queries..   I do have a pool of 20 parallel processes available in case a really large request comes along...so we
 do use all 20 parallel processes, but for the most part everything else gets downgraded to serial
 execution.

 On the single block read front, we have lots of upload jobs... about a month ago we switched to linux
 servers and they are much faster..which I'm assuming translates to more single block reads/waits..because these jobs lookup a single user in 2-3 tables, then insert/update another 2-3 tables..then repeat... and these run all day long. These jobs also need to lookup user information in an index that is used by everything in our system... and there is usually lots of waits on these datafiles... (this is expected and we are working on removing this bottleneck).

We have some users that also insist on updating/clearing columns on multi-million record tables during the day and this is where the large amounts of redo comes from ... Occasionally we generate almost 1MB of redo/sec..which to me is alot.

I guess the overall picture here is that we are continuing to grow very rapidly and with that comes more and more reading/writing of data. We have a plan to reduce IO by 30-40% ..which will be removing a very large index that is shared by the entire application..this will reduce io significantly.. I'll also look at high reads/writes in V$SQL.

I have thought about increasing my buffer cache by 25% for the next 2 months until we make some changes to reduce IO. Do you see any problems with this approach? Received on Tue Feb 28 2006 - 08:46:05 CST

Original text of this message

HOME | ASK QUESTION | ADD INFO | SEARCH | E-MAIL US