hfs versus vxfs for Cincom Supra and Oracle Databases

[Lisa Mauer]

Hi:
I know there are some posts out there that address some of this but not database specific.
Do companies still use hfs - besides us? We seem to be using hfs only for databases, why? I know this sounds like a stupid question, but is there an actual reason besides "It's always been that way because the hfs file system is better for databases"? We have numerous servers between us and our customers, too many to mention and everything from 10.20 to 11i.
Thanks
Lisa

[Clemens van Everdingen · Accepted solution]

Hi,

Since HFS is becoming unused and for performance issues vxfs is better. I would choose for vxfs.

Bill Hassel responded to this earlier today:

Most of the HFS documentation was taught in HP-UX advanced sysadmin courses years ago, but since HFS is no longer recommended as a production filesystem, there is very little in the way of details. It is the McKusick filesystem with the major pieces untouched for the most part. VxFS is the recommended filesystem due to overall performance and journaled logging. VxFS also provides numerous features not found in HFS such as online resizing and extensive mount options as well as a snapshot capability.

Hope this helps,

C.

[Jeff Schussele]

Hi Lisa,

I personally don't know of any companies using HFS for DB space.
I ALWAYS use vxfs & also prefer to create on an array of some sort. Mainly for the redundancy - both disk & path.
As it's a DB, dynamic inodes will not be a problem, BUT the journaling could save your rear-end in the event of a crash. And we ALL know that DBs *never* grow past their predicted size ;~) so the ability to extend the FS on-the-fly could keep the users smiling.
So tell your mgmnt to come on into the future & start using vxfs - they won't regret it.

My 2 cents,
Jeff

[A. Clay Stephenson]

Hi Lisa:

This is really one of those things that you need to measure for yourself but you are almost certainly going to find that vxfs (especially with OnlineJFS) will be at least 30% faster. Vxfs is really not sensitive to filesystem block size and if you are stripping, I think you will find that 64k is the optimum lvol stripe size. An Oracle blocksize of 8k tend to be a good all-around choice.

[Roger Baptiste]

hi,

I haven't seen any site which used HFS anymore (except for /stand which is a compulsion!).

I don't think, performance wise Vxfs is lagging anywhere behind HFS. It all depends on how you stripe and configure it. (8kb block for Database.) A big plus with Vxfs apart from it being standard is the integrity of the filesystem. In case the filesystem crashes, the fsck on it will come up very fast, since it used the logging concept for recording any changes to the Filesystem.

Then, you have the choice of resizing filesystem on the fly. You also have other options like mincache, convosync which is used for database filesystems. (to prevent buffering).

HTH
raj

[Lisa Mauer]

Thanks all for the information! It has been most helpful :)