topic Re: MinFree parameter on Journal file system VxFS in Operating System - HP-UX

MinFree parameter on Journal file system VxFS

Petr Drobny — Sat, 31 Jan 2004 09:20:51 GMT

I am going to go from HFS file system to journal VxFS file system. This file system have cca 10GB and cca 2.000.000 files/dirs.

1. Is there any sense try set MinFree on journal VxFS ?

2. If NO, How can I prevent file system to be FULL ?
I would like be sure, that the file system will be full to max 90%.

3. May I expect better performance ?
Nearly how many % ?

4. On my old HFS file system I have BLOCKSIZE/FRAGMENTSIZE 1024 Bytes - can be the same BLOCKSIZE on VxFS ?

Thank You very much for any information

Petr

Re: MinFree parameter on Journal file system VxFS

A. Clay Stephenson — Sat, 31 Jan 2004 12:12:28 GMT

1) There might be but there is no equivalent on a vxfs filesystem.

2) Just as you should be doing on any filesystem. By having daemons monitoring the situation. Note the minfree does not prevent applications from bombing out merely prevent a filesystem from filling up -- for non priviliged users.

3) Yes; it varies but 20% better is reasonable.

4) You can set bsize=1024 but unlike on hfs filesystems, blocksize has very, very minor impact upon performance. Vxfs filesystems are extent based.

Re: MinFree parameter on Journal file system VxFS

James Murtagh — Sat, 31 Jan 2004 12:13:30 GMT

Hi Petr,

You can't set minfree on a vxfs filesystem, hence you can't prevent the filesystem from filling up. I guess you will just need to proactively monitor it and take the appropriate action. I'm sure some of the other forum members have strategies they will share with you.....

Are you looking for performancce in general? As always, "it depends". The way vxfs is structured it performs optimally when large files are used as it takes advantage of extent allocation and also some advanced IO that reacts to different loads. I see you have about 2 million files - assuming an average number of blocks to files I would guess you probably won't see too much improvement, although only testing will confirm this. As to the block size you can set it at 1k, but the way it is structured in vxfs is much different to hfs. This is actually one of the things you'll need to be careful of - if you create a 10GB filesystem with the default options then vxfs will make the blocksize 8k - if what I have said above reflects your system you may find you run out of space quicker than you expect.

I have attached a document that describe a lot of these points, from the extent based allocation to IO tuning. Bear in mind this is for JFS 3.3, an optional package on 11.00 and the default on 11i, hence some of the commands such as vxtunefs will not be available on earlier versions. The latest version of vxfs on hpux is 3.5 and is available for 11i and 11.23.

http://www.docs.hp.com/hpux/onlinedocs/B3929-90011/B3929-90011.html

Cheers,

James.

Re: MinFree parameter on Journal file system VxFS

Bill Hassell — Sat, 31 Jan 2004 12:53:15 GMT

1) As mentioned, there is no concept of MINFREE on VxFS. Note that MINFREE on HFS was designed for HP-UX directories so that a full filesystem ona critical mountpoint would not prevent root from logging in. Unless you have auditing turned on, it usually is not an issue even with HFS. Having MINFREE does not prevent the filesystem from filling up! When a filesystem exceeds the MINFREE value, all applications and users see the filesystem as full. Only root sees the free space on the filesystem. Not much consolation since the applcation(s) are down.

2) It's real easy to prevent filesystems from filling up. DON'T allow any users to login!!! Seriously, the first step is to lock the directory structure (NEVER use 777 permissions for any directory in your data directories). Next is to write/rewrite the application(s) that can create files in this filesystem. The programmers must test that there is adequate room to create the new file and give a meaningful error message to the end user before attempting to clobber the filesystem.

If you have no control over the programs, you must monitor the filesystem through cron and shutdown the application when the filesystem nears 90% full, or at least notify someone that can take action.

3) Performance for VxFS is quite a bit better for HFS, but remember: filesystem design primarily affects creation/write speed, and to a lesser extent, open speed. Read speeds are basically the same for all filesystems and are influenced much more by LVM layout (extent striping over many disks for random access).

4) You can ignore the BLOCK/FRAG values as they don't work the same way as HFS. VxFS does not have a fixed set of inodes as they are created as needed. bdf -i is basically meaningless for the inode free metric.

Re: MinFree parameter on Journal file system VxFS

Petr Drobny — Sat, 31 Jan 2004 15:15:15 GMT

Your answers are excellent

Thanks to ALL

Petr

(I am sorry for my English)

Re: MinFree parameter on Journal file system VxFS

Jeff Schussele — Sat, 31 Jan 2004 18:38:35 GMT

You gotta love that - YEP - you can't fill a FS if you don't allow users to login.....
Bill - you've obviously too much time on your hands since you've retired...

Respectfully,
Jeff