krish, the CPU utilization never goes up more than 30%. and luckily no log/trace files on all stack OS/clusterware/database, when reboot happened. MTS is another plan.... but why there is so many big difference between oracle reported (AWR) and OS?? I thinking of OS memory management problem, how to dig into it? oh ya OS version is B.11.11 HP-UX PA-RISC. regards ujang On Jan 20, 2008 8:20 AM, <krish.hariharan@xxxxxxxxxxxx> wrote: > Ujang, > > Since you mention RAC I suspect this is probably node eviction - the cause > could be cpu utilization (not mentioned) or swap consumption preventing > heart beat messages and consequently resulting in node eviction. Look at the > ocssd.log on the surviving node. How many nodes in your cluster and how busy > are the other nodes (cpu, swap)? > > If my understanding of memory usage is correct (as seen in Solaris) the > process maps the SGA and is part of its virtual space and has to reserve > swap proportional to it. I address that using shared servers (MTS) since > only a few processes (in your case 50) need be active and not the 1500 since > you can get away with about 75 shared servers processes (if you are not > already using this capability). > > The system logs or the crs/css logs will give you that information. The > other thing to look for is the progressive build up and usage of swap. > > Regards, > -Krish > Krish Hariharan > President/Executive Architect, Quasar Database Technologies, LLC > http://www.linkedin.com/in/quasardb > > > -----Original Message----- > From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx] > On Behalf Of Ujang Jaenudin > Sent: Saturday, January 19, 2008 5:36 PM > To: oracle-l > Subject: memory problem - oracle 10203 on HPUX > > folks, > > I have a machine (RAC configuration) 10.2.0.4 on HPUX - PARISC. > in a node I have 38GB physical memory, 8 processors. > the machine could not run with asynchronous IO, so I set io slave to > 8, and dbwr to 1. > > at AWR report, it said that 2GB memory for PGA is enough, and 12GB of > SGA mostly enough. > concurrent user is 1000 until maximum 1500 users connected. > at one time only 30-50 concurrent active with OLTP transaction. > > at the OS perspective, it said that mostly 90-100% memory has been used, > even when high load, 50% of the swap will be used also. > almost every 20 days, the server is rebooted itself, > I think this is due to lack of memory.... > > why between OS and oracle side there has much big different of memory > usage??? > > -- //www.freelists.org/webpage/oracle-l