vxfs filesystem fragmentation

[Adam Garsha]

1. It looks like the way to check fragmentation is "fsadm -F vxfs -E /filesystem", correct?

2. Is it ok to check fragementation while filesystem is live in production (is there any noticible performance impact)?

3. What is consider "Bad" fragmentation by your favorite rule of thumb?

[Pete Randall]

Adam,

I run this once a week via cron:


for i in `bdf -l |grep -v Filesystem |awk '{ print $6 }'`
do
if [ $i != "/stand" ]; then
echo "Re-Org'ing " $i
fsadm -F vxfs -d -D -e -E $i
fi
done


This adds directory reorganization to the extent reorganization you specified. Cron'ing it in the middle of the night avoids any negative impact. Running it once a week ensures that I will not have "Bad" fragmentation.


Pete

[Peter Godron · Accepted solution]

Adam,
1. Correct Syntax

2. Will impact on performance, run out-of-hours if possible

3. See http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=86624

Another way to measure: Just before the user notices it.

[Adam Garsha]

I will hold points until I get a few responses.

[Geoff Wild]

Here's what I run from cron weekly:

#defrag filesystems
0 22 * * 4 /usr/local/bin/fsadm.defrag >/dev/null 2>&1


# cat /usr/local/bin/fsadm.defrag
#!/bin/sh
# defrag all mounted file systems
LOG=/tmp/fsadm.defrag.log
if [ -f $LOG ]
then
mv $LOG $LOG.old
fi
cat /dev/null > $LOG

for i in `mount -l | grep -v stand |awk '{print $1}'`
do
echo "defraging " $i "at "`date` >> $LOG
fsadm -F vxfs -d -D -e -E $i >> $LOG
done


Rgds...Geoff

[A. Clay Stephenson]

In almost all cases, don't expect very much in the way of performance improvements from this. Vxfs filesystems are already extent-based which means that they try to allocate large chunks of a file. Only in exceptional cases, have I ever seen improvements more than 1-3% even for what appears to be a highly fragmented vxfs filesystem. In real life, running fsadm once every 3 months or so will be almost as good as weekly.

[Steven E. Protter]

Shalom,

I didn't see it mentioned that Online JFS is a requirement for this.

I found that running this caused my SAN admins headaches with high I/O, so schedule it for a time that load is light.

A. Clay is always right, the performance payback isn't real huge, but every little bit helps.

SEP

[Adam Garsha]

Is there any performance impact just to look, but not to fix?

[A. Clay Stephenson]

There is, of course, some performance impact but it is small --- but if you are looking for significant performance gains then look elsewhere.

[Adam Garsha]

Yes, I don't think fragmentation is a problem, but I need to convince some folks.

I haven't seen fragmentation being a problem on any modern unix filesystem. (That's why I know so little about defraging vxfs... because my knee-jerk reaction is always "that ain't the problem")

[Bill Hassell]

> Yes, I don't think fragmentation is a problem, but I need to convince some folks.

This *ALWAYS* happens because 'some folks' only know one kind of OS and see everything else with the same viewpoint. Sort of like: If you only have a hammer in your hand then all your problems tend to look like nails.

One of the most common requests for defragmentation comes from authors of lousy SQL code.

[Adam Garsha]

above.