Re: How do I use raw devices?

From: Lon Culbertson <culberts_at_netcom.com>
Date: Tue, 22 Mar 1994 07:48:33 GMT
Message-ID: <culbertsCn230y.6q0_at_netcom.com>


jl34778_at_corp02.d51.lilly.com wrote:
: I am running ORACLE 7.0.15.4 on a SparcServer 1000, running Solaris 2.3. I am
: beginning to look into using raw devices, and wondered if anyone has any advice
: for me.
 

: The two reasons why I'd like to use them are:
 

: a. To decrease the annoying 20% space overhead that seems to come with
: formating a drive and putting a file system on it.

    Well, 6.25 % of the overhead comes about when newfs is used with the default 2048 bytes per inode, which creates 488,000+ inodes on a 1GB disk. Since each inode uses 128 bytes, one has just successfully (as it were) used approximately 62.5 MB of one's disk space.

    Assuming that one is using a disk only for Oracle datafiles, one can reduce inode overhead to approximately 4 MB for a 1 GB disk by using newfs -i 32768, which will create approximately 32 K inodes. This many inodes is still overkill, so another 3.5 MB could be saved by using -i 262144, which would still result in 4 K inodes on a 1GB disk, I believe.

    Another 10% of the overhead is due to the newfs default of reserving 10% of disk space. Again, this can be eliminated by using the appropriate nondefault parameter.

    While one is at it, increase the size of the cylinders per cyclinder group, with something like -c 64.

    Then, after the newfs has been done, look at tunefs, to adjust the maximum bytes per cylinder group (?).

: b. To increase I/O performance.
 

: My questions are:
 

: However, the tablespace files will vary in size, so the partitions will need
: to accomodate that. What does it do to performance, etc., if you allocate
: 400m to a tablespace using 8 50m files?

    All that I know about raw partitions is what I've read, but I believe that one can only have a one-to-one correspondence of datafiles to partitions.

: 2. How many partitions can you have on a disk?

    Eight. Received on Tue Mar 22 1994 - 08:48:33 CET

Original text of this message