definition of f_bsize vs. f_frsize

[John Jimenez]

Over the years as file systems have grow to 300 and 900 gigs I have increased my pe size from 4 to 16 and now 32. But I have never modified block size, I show frsize to be 1k but bsize to be 8 k. Other threads say that frsize it the block size. If this is true what is bsize then? can you tell me the difference? Here is the out put from fstyp command. Also thinking what is best to set Oracle 10g database at?


maxux1:[/]
# fstyp -v /dev/vg19/lvol19
vxfs
version: 5
f_bsize: 8192
f_frsize: 1024
f_blocks: 733806592
f_bfree: 729478676
f_bavail: 683886259
f_files: 182375844
f_ffree: 182369668
f_favail: 182369668
f_fsid: 1074987009
f_basetype: vxfs
f_namemax: 254
f_magic: a501fcf5
f_featurebits: 0
f_flag: 16
f_fsindex: 9
f_size: 733806592

[John Jimenez]

one post says frsize is block size

http://h30499.www3.hp.com/t5/System-Administration/Filesystem-block-size-fstyp-v-vs-mkfs-m/m-p/2577765#M31405

But a guy on this post says that bsize is block size.

http://h30499.www3.hp.com/t5/System-Administration/how-to-know-the-block-size-and-chunk-size-of-my-hp-ux-system/m-p/4264214#M333302

You would think bsize would be block size, but you see many on the first thread say that its frsize. which one is correct? The MAN pages in this case is useless.

[John Jimenez]

I am using jfs (not online jfs) I usually use the command line for the vgcreate because I have to modify extents size and numbers. But I was using SAM before because I was only enabling large files and using other defaults. But in SAM I do not see an option to modify block size. so maybe time to be a big boy and start using lvcreate and newfs commands. I am reading a book to and is says that block size can only be modified on hfs....not jfs. Does that sound right?

[Jose Mosquera]

Hi,

For your first doubt please check:
http://www.docs.hp.com/en/B2355-90682/statvfs.2.html
-Or-
#man statvfs

If you are on 11i then you have option to modify PE size at vgcreate command:
#vgcreate -s [nn] vgname /dev/dsk/cntndn...
Where -s [nn] will be your desired PE size.
#man vgcreate

If you're over 11i (11i v2 with the PHCO_35524 patch), you could resize disk online with vgmodify. Please check this:
http://h20000.www2.hp.com/bc/docs/support/SupportManual/c01920387/c01920387.pdf

Rgds.

[John Jimenez]

Yes, for years as LUNS have been growing, I have already modify the PE size from 4 to 8 and now 16 and 32 and also increasing the ammounts. But I was not asking about PE, I asking about modifying the "block" size on jfs during the newfs

[John Jimenez]

We are doing Oracle 10g Exports on a RAC clustered system on HP_UX 11.23 system using Raid 1/0 on EMC CLARiiON CX500. It was taking 15-16 hours to send it to SATA disks on Raid 1/0 of that same CLARiiON. I created a 300 GIG disks on RAID5 LUN on the fiber. It was a few hours faster, but still took almost 12 hours. I think they have an option to set the block size when they do the export. They would prefer it to be 8k blocks, but I was reviewing the newfs command and saw anot the the -b only works with HFS. Why would that enhanced functionality not be on JFS, but it is on older HFS?

[Jose Mosquera]

Hi John,

Please check these articles about "Oracle and RAID Usage" and "Setting the optimal disk RAID stripe size", are easy to read, interesting and useful:
http://www.dba-oracle.com/oracle_tips_raid_usage.htm
http://www.dba-oracle.com/t_setting_optimal_raid_asm_disk_stripe_size.htm

I hope they are in your profit!

[Matti_Kurkela]

frsize is the "fundamental block size": the size of the smallest independently addressable/allocatable data block on the storage.

bsize is the "preferred block size": the system will attempt to do all operations in multiples of this size unless it absolutely must do otherwise.

I would expect bsize to be always equal or greater than frsize.

It may be possible to change bsize on an existing filesystem, but changing frsize will probably always require backing up all data on the filesystem and re-running mkfs.

If you're using SSD storage for maximum performance, the value of bsize may be important: SSD blocks must be erased before they can be re-written, and the erase operation will always erase an entire "erase block", which is usually greater than 1k in size. To optimize the usage of SSD, you'll want bsize to be equal to the SSD erase block size (or an exact multiple of it). Otherwise, two write operations to consecutive disk blocks may cause the same erase block to be erased and re-written twice, which lowers performance and speeds up SSD wear rate.

MK

[John Jimenez]

Jose, Thanks for the links. I already understand the raid types. All our oracle database is on raid 1/0 on EMC CLARiiON 500. It is also on ASM RAC on two RP7420 servers. My issues is with the export taking to long. I was doing them to five SATA raid5....purchased more SATA drives and chaned it to Raid 1/0 running on 8 SATA drives and it improved about 10%, but still takes 16 hours. Now I moved it to Raid5 on five fiber 15k drives and shaved off another 3 hours, but 12 hours still seems like a long time. I have to move some things around and will next try to export to RAID 1/0 on 12 fiber drives. I really think my issue is EMC CLARiiON 500 might be too slow.

[John Jimenez]

Thank you for that explanation. So when Oracle admins ask what the block size is, it sounds like I can say that it is 8k. right? I have fiber 15k and SATA. I wish I had SSD, maybe then this 300 gig export would not take a 1/2 day. I am gong on vacation in the middle of the week. I am going to see if any other helpful hints come in and will assign points on Monday. Thanks again for that explaination on b and fr sizes.