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: Understanding Oracle blocks and UNIX blocks re. redos

Re: Understanding Oracle blocks and UNIX blocks re. redos

From: Dave <aixman_NOSPAM__at_flash.net>
Date: 2000/03/25
Message-ID: <38dcabc2_3@news1.prserv.net>#1/1

Joseph Blazic wrote in message <38DAC94D.1F43DC1E_at_atu.com.au>...
>Hi,
>
>Can someone please explain the relationship between Oracle blocks and
>UNIX blocks?
>Say you set block size in initora to 2048.
>When Oracle goes to read a file, does it request of UNIX "x" blocks to
>be read, or "y" bytes?
>Same question when Oracle writes to a file...
>
>I ask this so I can also understand the implication of using
>large-file-enabled filesytems (feature of AIX) with Oracle. For
>filesystems containing:
>1. Data
>2. Oracle dump directories and redo logs.
>
>Thanks and Regards,
>Joseph Blazic
>

  Well, I'm not so sure about "UNIX" blocks, but I can semi-competantly discuss block sizes in AIX, for what it's worth.

  Basically, if you are using a JFS filesystem on an AIX LVM supported disk subsystem, then your I/O block size is 4096 bytes (4K). This just happens to match exactly the "frame" and "page" allocation size in the AIX virtual memory manager (VMM), primarily because almost all JFS related I/O is handled by VMM in AIX. Also note that the maximum logical I/O request that the logical volume device driver can service is 128K (32 4K pages), anything larger is broken into multiple logical requests. So, having a block size larger than 128K is probably not going to give you any advantages in performance.

  In a nutshell, oracle will hand X number of bytes to read/write to AIX through a read/write call, which in turn passes the request to VMM for processing. Within VMM, persistent storage segments are allocated and 4K pages
are read from the filesystem (if not already there) into the appropriate free 4K memory frames for use. If an update occurs to a given page, then it is marked for update during the next sync of that segments modified pages.

  If you are concerned about performance, then your best performance will most likely be achieved when you set your Oracle "block" size to match what the AIX VMM uses, either directly (4K) or even multiples thereof (when using JFS filesystems). This would be true regardless whether they are large file enabled or not.

   Where the large file enabled filesystem may impact you most is with the allocation of space, especially if Oracle is using sparse allocated files for it's database files (which I believe it does). This could cause you problems in that your large file enabled filesystems may have to be considerably larger than with "regular" filesystems, just to provide enough disk "blocks" to accomodate large sparse files.

   Hope this helps. Received on Sat Mar 25 2000 - 00:00:00 CST

Original text of this message

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