0% found this document useful (0 votes)
125 views2 pages

IBM AIX RISC System 6000 Error

210385

Uploaded by

Atul Sawant
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
0% found this document useful (0 votes)
125 views2 pages

IBM AIX RISC System 6000 Error

210385

Uploaded by

Atul Sawant
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF or read online on Scribd
Sy SAPNote 210385 - AIX 4.3.3: error: 11: Resource temporarily unavailable Component: BC-DB-ORA (Oracle), Version: 5, Released On: 22.06.2000 | symptom error=27063 txt 'IBM AIX RISC System/6000 Error: 11: Resource terporarily unavailable, Oracle database shuts down. AIX 4.3.3 and Oracle using asynchio. DB using 2 small number of Targe disks and asynchia paraneters have not been tuned or not sufficiently tuned. | other Terms AIX 4.3.3, Oracle, asynchio, performance, error=27063, error 1242 | Reason and Prerequisites Increase the number of asynchio requests which can be queued (maxreqs) and reconsider the To subsystem layout ( use more files rather than large files if possible, access the number of spindles available). The errno 11 is because the ATO maxregs parameter is being hit (defaults to 4096). Most Vikely you'll see this in 4.3.3 and newer releases. Prior to 4.3.3, when the dbwriter would issue a call to async 1/0, it could get blocked on the inode lock while in the ATO kernel extension but before the I/os were put on the ATO queue. Because of that, not as many T/os could be on the queue However, the real behavior should be that the dbwriter should never get blocked on the inode Jock because it's the AIO kprocs that are doing the writes to the files Cin the case of 3FS databases) The dbwriter should just do it's lio_listio() to put the Aros on the queue and return immediately (that's the whole purpose of AIO, to submit the I/os and return immediately). So in 4.3.3, AIO was changed so that the inode lock is not taken by the process issuing the I/os. It is taken when the kprocs do the write. Previously, the lock was taken by both. Because of this change, the dbwriter is now free to submit a lot more I/o requests. Now it can start flushing out memory very quickly - but unfortunately, it can now flood the Aro subsystem with so many I/os that it can fill up the ATO queues (which can hold up to maxreq requests). once maxreq is hit, error 11 is returned. If there are only a few files, then there will be a lot more inode lock contention. This means that it will take longer for the ATO kprocs to do their work. In that case, it will be harder to keep up with the requests by the dbwriter. So it could help to increase the number of ATO servers in case some I/Os go to other files. aut maxreqs should definately be increased. (You could start by doubling or tripling the numbers). If using just a few disks instead of many spindles, then at least the files should be split up into many files to alleviate the inode lock contention. The number of outstanding requests for a single file is limited only by maxreqs as long as the VM 1/0 pacing is not active (rinpout/maxpout=0) . 1/Os may also get delayed if you run out of bufstructs. Tf you have at least bos.adt.samples.4.3.3.10, then you can run /usr/samples/kernel/vmtune -a to see if you are running out of bufstructs. If fsbufwaitcnt is non zero, and keeps increasing, ‘then vmtune -b value needs to be increased. ay default, you only get 93 bufstructs per filesystem. The modified value only goes into effect when the filesystem is mounted, so you will need to unmount and then remount the filesysten to get more active bufstructs. The number of outstanding 1/os for a single disk depends on the queue depth of the disk and also how many T/os the adapter can hold. UPDATE: The following input received from a successful problem solution in the field ATO maxregs = 12288 und Jusr/samples/kernel/vmtune -b 200 (numfsbufs von 93 auf 200 erhdht) The initstD.ora parameter for asynchTo is incorrect in several notes: it should be: disk_asynch_io =true (or false accordingly) | Software Components Software Component From To ‘And subsequent SAP_APPL 310 ail SAP_APPL 404 404, SAP_APPL 454 458 SAP_APPI. 468 468 I This document refers to SAP Note/KBA Component Tide 79313 BC-DB-ORA FAQ: Oracle 1/0 configuration 546006 BC-DB-ORA, ‘Problems with Oracle due to operating system errors 328785 ORA-00376: File cannot be read at this time | This document is referenced by SAP Note/KBA Component Title ‘EAQ: Oracle 1/O configuration 546006 BC-DB-ORA Problems with Oracle due to operating system errors 328785 BC-DB-ORA, ORA-00376: File eannot be read at this time

You might also like