Hi Mauro, Would that also be the case in 2.6 kernels?
Anyway, we already have direct_io enabled. Back in 2006 we had only async_IO and hit a kernel bug because of it. After direct_io was activated as well the kernel bug didn't happen again. (I think the IO load was so high that the async buffer wrapped. Unfortunately we weren't able or allowed to prove that.) Actually early last year you had mentioned this example. So I tried to convince them of 'the mainframe way'. Let's not go into details on their reaction but let's say it was less than professional. The result was an increase in memory because that's the only way you can solve performance issues. (Just like they didn't believe the kernel bug was caused by an IO load of over 25K/sec, because there is no way any computer can achieve such high IO loads.) Good link. I'll discuss this with the oracle guy to see what the current settings are. Most links I had found so far regard performance problems only from the runtime view. So increase parameters to speed up the run but don't explain what the memory impact is. After all, we have plenty of cheap memory, right? Thanks, Berry. -----Original Message----- From: Linux on 390 Port [mailto:[email protected]] On Behalf Of Mauro Souza Sent: Wednesday, October 02, 2013 6:46 PM To: [email protected] Subject: Re: Oracle RMAN OOM I heard that activating DIRECT_IO helps solve the RMAN memory eating problem on kernels 2.4. On http://dbasolutions.wikispaces.com/RMAN+Performance+Tuning you can see some ideas on how to tune rman memory usage. To show your Oracle DBA's that more memory is less performance, ask them to try the "Mainframe Way". Get some heavy queries, get them to run them using the current configuration. Run it 3 or 4 times. Get the average running time. Change a couple parameters: - Use SGA max size = 66% of available memory - Use 4GB on the guest - Use DIRECT_IO - Use at least 40 io slaves Run again 4 times. Compare the results. The last time I did this, we got a 39 min job running in 9 minutes. The DBA laughed on me, saying I crashed Oracle and the queries died. He opened the results table, everything was correct. He got angry, started again the queries. 9 minutes. He did it again, 9 minutes again. And I spent the next hour explaining why a mainframe is not a x86 box... Mauro http://mauro.limeiratem.com - registered Linux User: 294521 Scripture is both history, and a love letter from God. ---------------------------------------------------------------------- For LINUX-390 subscribe / signoff / archive access instructions, send email to [email protected] with the message: INFO LINUX-390 or visit http://www.marist.edu/htbin/wlvindex?LINUX-390 ---------------------------------------------------------------------- For more information on Linux on System z, visit http://wiki.linuxvm.org/
