LVM and VxVM
cancel
Showing results for 
Search instead for 
Did you mean: 

What controls the f_frsize parameter you see in the fstype -v output

 
Mel Burslan
Honored Contributor

What controls the f_frsize parameter you see in the fstype -v output

I have exhausted all the man pages I could find and read (in my opinion of course) and none referenced how to specify this parameter. In the case it is not directly addressable with the commands, I'd like to know the addressable/changeable parameters in any command to make this value different.

 

My situation:

On the same hpux 11.23 operated rx6600 server I have this oracle db underlying a SAP instance

all filesystems mounted on /oracle/blah/sapdata1, 2, 3 etc, when run against fstpe -v command gives this o/p:

 

# fstyp -v /dev/vgSID/lvsapdataX # x=1-14
vxfs
version: 6
f_bsize: 8192
f_frsize: 1024
f_blocks: 266338304
f_bfree: 56542711
f_bavail: 53008792
f_files: 14135708
f_ffree: 14135676
f_favail: 14135676
f_fsid: 1081737240
f_basetype: vxfs
f_namemax: 254
f_magic: a501fcf5
f_featurebits: 0
f_flag: 16
f_fsindex: 10
f_fsize: 266338304

And they wanted a dump directory, which is a 3TB single lvol and when I run the same fstyp command on this volume, this is what I am getting:

 

# fstyp -v /dev/vgSIDhuge/lvSIDdump
vxfs
version: 6
f_bsize: 8192
f_frsize: 8192
f_blocks: 379994112
f_bfree: 379961055
f_bavail: 376992610
f_files: 94991296
f_ffree: 94990240
f_favail: 94990240
f_fsid: 1081147393
f_basetype: vxfs
f_namemax: 254
f_magic: a501fcf5
f_featurebits: 0
f_flag: 16
f_fsindex: 10
f_fsize: 379994112

And when dbas performed an operation called datapump (don't know if it is something they come up with or is it a legitimate oracle term) the data transfer rate is abysmal at best. They said the f_frsize is the culprit as the system needs to make 8 system calls to read data to be written to the dump dir only once and it is taxing the performance. I don't know how much of this statement true but I can see the different f_frsize settings. Just to prove them right or wrong, I want to create a new fs with 8192 f_frsize and have them use it as the source fs and see if it helps anything.

 

How can I do that ? or is there something more beneficial to do here ?

 

Thanks.

________________________________
UNIX because I majored in cryptology...
4 REPLIES 4
James R. Ferguson
Acclaimed Contributor

Re: What controls the f_frsize parameter you see in the fstype -v output

Hi Mel:

 

A value if chosen by default when you do a 'mkfs_vxfs'.  From the 11.23 manpages for 'mkfs_vxfs' [which do NOT appear in the 11.31 online ones for some unknown reason]:

 

/* begin quote */

bsize=bsize bsize is the block size for files on the file system and represents the smallest amount of disk space allocated to a file. The value of bsize must be based on a power of 2 selected from the range 1024 bytes to 8192 bytes. The default is 1024 bytes for file systems smaller than two terabytes. For Version 5 disk layout file systems larger than two terabytes, the maximum file system size is dependent on the block size. See the "VxFS Version 5 Disk Layout File System Sizes" topic in this section for more information. If bsize is not specified, the block size defaults are set to the appropri- ate value, based on size, when a file system is created.

/* endof quote */

 

I hope this helps.

 

Regards!

 

...JRF...

Mel Burslan
Honored Contributor

Re: What controls the f_frsize parameter you see in the fstype -v output

m,

 

Thanks for the insight. Let me tell you a bit more as I unravel the things since my original post

 

first off : both sapdata volumes and my 3TB volume reside on the same Hitachi SAN Array (NOT HP's XP but really Hitachi labeled SAN Array)

 

my original sapdata volumes are on a VG, all PVs of which has 2 paths to the SAN

my 3TB volume has different size PVs, all over the map size-wise and they all have 3 paths to SAN

I am not involved in SAN allocation and don't know why it is this way

 

my sapdata volumes are less than 300 GB each and fstyp reports their f_bsize at 8K but their f_frsize is reporting 1K

From what I have read 8K block sizes is what we are interested in 99.99% of the time because as long as it is allocate at 8K blocksizes both read and write operations will be performed at these boundaries. frsize only comes into effect when there is not enough data or space to allocate the fuill 8K block size. Is this a correct assumption ?

 

Also, is there a significant effect of this f_frsize value, when it comes to how oracle handles the disk i/o, let it be read or write activity ?

 

Thanks again

 

Mel

________________________________
UNIX because I majored in cryptology...
James R. Ferguson
Acclaimed Contributor

Re: What controls the f_frsize parameter you see in the fstype -v output

Hi (again) Mel:

 

Perhaps too, this TKB document helps distinguish the 'bsize' (largest possible block size) from the 'frsize' (fragment size):

 

Document ID: emr_na-c00934933-2

You can find this document by logging into the HPSC ( http://www.hp.com/go/hpsc ) and clicking on the Technical Knowledge Database link on the left.  At least that should look familar :-)

 

Regards!

 

...JRF...

donna hofmeister
Trusted Contributor

Re: What controls the f_frsize parameter you see in the fstype -v output