Re: Filesystem Block Size ? 1K or 4K or 8K ?
- From: "Tim Gorman" <tim@xxxxxxxxx>
- To: oracle-l@xxxxxxxxxxxxx
- Date: Tue, 14 Oct 2008 21:16:32 +0000
This link from HP should help
(http://h71019.www7.hp.com/ERC/downloads/4AA1-9839ENW.pdf)...
Paraphrasing from this article...
Redo log file systemsCreate redo log file systems with a 1 KB block size; use
the following mount options:
delaylog,nodatainlog,convosync=direct,mincache=direct
Archive log file systemsCreate archive log file systems witha 1 KB block size.
Mount options vary depending on the particularversion of JFS, as explained
below.
JFS 3.3 and 3.5, without the sparse file I/O patch
For JFS 3.3 and 3.5 without the sparse file I/O patch (11.11: PHKL_32355,
11.23: PHKL_34179), use the following mount options:
delaylog,nodatainlog
Modify the /etc/vx/tunefstab file,changing the discovered_direct_iosz parameter
for the archive filesystem to 2097152, which enhances performance when writing
archivefiles.
Note: Archive files are sparse files, which perform better with buffered I/O
JFS 3.5, with the sparse file I/O patch
For JFS 3.5 with the sparse file I/O patch, use the following options to mount
the archive directory for direct I/O:
delaylog,nodatainlog,convosync=direct,mincache=direct
JFS 4.1 on HP-UX 11.23 and 11.31
The sparse file I/O issue is hasbeen resolved in JFS 4.1. Use the following
options to mount thearchive directory for direct I/O:
delaylog,nodatainlog,convosync=direct,mincache=direct
Oracle tablespace file systemsHP offers the following suggestions for creating
Oracle tablespace file systems:
Using HP-UX buffer cache
Create file systems with an 8 KB block size; use the following mount options:
delaylog, nodatainlog
While using HP-UX buffer cache may providesome benefit through JFS read-ahead,
it may be better to bypass buffercache in environments that are very
I/O-intensive.
Bypassing HP-UX buffer cache
Direct I/O can be more efficientthan using buffer cache and provides some
benefit in I/O-intensiveenvironments. Note, however, that it is critical to set
the Oracledb_file_multiblock_read_count value to 32 to avoid multi-block
I/Osbeing broken down into individual, single-block, physical reads
(seeSequential I/O penalty with readv and Higher Oracle multi-block readcount
values).
Create file systems with an 8 KB block size; use the following mount options:
delaylog,nodatainlog,convosync=direct,mincache=direct
Oracle 10g and 11g environments
In an Oracle 10g or 11genvironment, allow the value of the
Oracledb_file_multiblock_read_count parameter to remain at default,
whichresults in1 MB reads that have no impact on the Oracle Optimizer logic.
Use the following options to mount a file system and specify direct I/O:
delaylog,nodatainlog,convosync=direct,mincache=direct
Oracle binaries file systemsUse the following mount options:
delaylog,nodatainlog
Pretty much says it all. Any questions?
Hope this helps...
Tim Gorman
consultant - Evergreen Database Technologies, Inc.
P.O. Box 630791, Highlands Ranch CO 80163-0791
website = http://www.EvDBT.com/
email = Tim@xxxxxxxxx
mobile = +1-303-885-4526
fax = +1-303-484-3608
Yahoo IM = tim_evdbt
VIVEK_SHARMA wrote:Folks
We are in the processof Setting up DB Server for an OLTP Banking Application
Benchchmark.
HP-UX B.11.31
system device blocksize(think this is the O.S. block size) = 1K
Base page size = 4K.
DB_block_size = 8K
Should the Filesystem(VXFS) Block Size be set to 1K, 4K or 8K? …Personally I
think it should be 4 K (i.e. Equal to the Page-size)
Cheers & Thanks Vmuch
Vivek
P.S. Will provide anydata needed
Other related posts:
