Re: Oracle Data Warehousing, UNIX and large file-enabled file systems

Don..

You should, if you can, get something like Veritas F/S... Then you would have the option of mounting the F/S bypassing the Kernel Buffer Cache...
Also true JFS handles inodes & extents dynamically, so this would virtually eliminate the System CPU overhead to maintain superblock, inodes..etc..Also, mounting F/S this way equals or betters Raw Logical volumes..As well, if you can change the stripe size to the Kernel value scsi_phys_max for your LUNs, I am sure that will help as well..and by turning on the large files option, you need not ever worry about the F/S fragmentation....
I am at an HP-UX shop hosting OLTP instances (JFS 3.3, 64-bit rdbms 8.1.6.2 &9.0.1.1,HP-UX 11.00 64-bit, XP-256 Disk Farms) and was able to prove all this..

HTH

--
Zongk Tu

Oracle Apps DBA/Unix SA

Don Gillespie wrote:



> I am the DBA for a data warehouse environment that is expected to get

> to about 3TB.  That would mean about 1500 or so data files with the

> 2GB file limit.  Besides being a nightmare to manage that many files,

> I anticipate the overhead on checkpoints would be tremendous.  The

> environment is 32-bit Oracle (possibility of 64 bit in the future) on

> AIX with disk storage on an IBM Shark SAN (RAID5, 32K stripe; no

> choice here), with a 16K Oracle block size (the max allowed).  We are

> using Journaled File Systems, not raw partitions.  I am contemplating

> the use of large file-enabled JFSs for all JFSs that would contain

> oracle data files, log files and control files.  But I don't know much

> about them, and I am wondering if there are serious performance, space

> consumption or administration issues in doing so.

>

> I understand with large file-enabled JFSs, I could have 64GB files

> (one posting says 32GB is max for Oracle) and up to 8TB JFSs.  But I

> have heard conflicting comments on implications for INODEs, allocation

> units and fragmentation sizes(?).  Am I going to suffer an I/O penalty

> every time the database attempts to write blocks out to the files,

> especially if the blocks are non-contiguous?  What about disk space

> usage?  Any other issues (backup/recovery, RMAN, EXP/IMP, etc.)?

>

> I would be particularly interested in what others have experienced.

> What file sizes did you allow?  What size of JFSs were allowed?  What

> issues were encountered?  If you had the chance to start from scratch

> again, would you go large file-enabled again, and if so, what would

> you set as your maximum file size and JFS size?

>

> These issues have come to the forefront as I prepare to create the

> tablespaces, so unfortunately this is a time-sensitive issue.  Any

> input would be GREATLY appreciated!


--
Zongk Tu

Oracle Apps DBA/Unix SA