Re: Memory leak

On Wed, 24 Sep 2003 22:31:48 +0200, "Langelage, Frank" <frank_at_lafr.de> wrote:

>Sybrand Bakker wrote:
>> On Wed, 24 Sep 2003 09:54:15 +0200, "Langelage, Frank" <frank_at_lafr.de>
>> wrote:
>>
>>
>>>Anna C. Dent wrote:
>>>
>>>>Adolfo wrote:
>>>>
>>>>
>>>>>We check with top and sar -r
>>>>
>>>>
>>>>You do NOT have a memory leak.
>>>>This is typical behavior for Oracle on Solaris.
>>>>The "lost" memory is being used by Solaris for
>>>>file system buffer cache.
>>>>
>>>
>>>Don't know about Solaris 2.6, but on Solaris 8 or 9 you can mount
>>>filesystems with option "forcedirectio".
>>>Do this only for filesystems which contain oracle datafiles only,
>>>nothing else.
>>>This flag avoids the buffering of the oracle datafiles by the os. The
>>>buffering is not needed, because oracle buffers data and index. So you
>>>eliminate one memory transfer which also consumes some time.
>>
>>
>>
>> forcedirectio is available for ufs filesystems only. Using ufs
>> filesystems for Oracle can not be recommended.
>>
>
>Why not ? Can you explain this a little bit more ?
>What's the recommended filesystem type for Oracle on Solaris then ?
>Veritas-FS (VxFS) ?

ufs is pretty old technology, and it is buffered. The way Sun implemented the filesystem cache on Solaris, there is not much distinction between virtual memory available to programs and virtual memory available to the filesystem cache. They are both implemented in virtual memory and both can be adjusted dynamically. On Unix you must always discriminate between directio and async io.
They are not the same. Directio means basically bypassing the buffer cache. Nothing more, nothing less. Asynchronous I/O is not possible on ufs. You'll need raw volumes for that, so volumes without a filesystem. Obviously, they are extreemly unmanageable: you won't be able to extend a tablespace as the tablespace maps to a raw disk volume.
That's why quick I/O and Veritas were invented. There are several comparative studies online on http://docs.sun.com and http://www.sunsolve.com Unfortunately I don't remember the complete contents, and I don't have the URL handy.

Regards

Sybrand Bakker, Senior Oracle DBA

To reply remove -verwijderdit from my e-mail address Received on Thu Sep 25 2003 - 12:27:31 CDT