Help with not enough free physical extents and lvcreate error msg

[Doug_3]

Hello, I have an fc10 jbod of 10 drives 18 gig each. One yr ago I created a pv mirrored and extent based striped. I did not specify the extent size in the vgcreate (oops!!!).

I have added 25 lvols over the year with success. According to SAM I still have quite a bit of space available.

However, this time I generated a error:
lvcreate -D y -s g -m 2 -L 2048 -n fcdbstest6 /dev/vg05
Logical volume "/dev/vg05/fcdbstest6" has been successfully created with
character device "/dev/vg05/rfcdbstest6".
lvcreate: Not enough free physical extents available.
Logical volume "/dev/vg05/fcdbstest6" could not be extended.

I tried searching the forums but am unsure how to remedy the problem. Is the issue the extents on the pvg or ???? Can I use an lvextend for this specific issue and what can I do in the future. I am sure I'll need to add new lvols as time goes by.

Thanks in advance,
Doug

[James R. Ferguson]

Hi Doug:

It appears that you do not have enough free physical extents to satisfy the constaint imposed by PVG-strict allocation with mirroring. See the man pages for 'lvcreate' and 'lvextend' for more information.

Regards!

...JRF...

[Marco Santerre]

Doug, it would be interesting to have a display of your Volume group here so that we have an idea what is happening in there, and how your disks are used, etc..

though you are speciafying to have two mirror copies of this LV, is that right?

[Patrick Wallek · Accepted solution]

It souds as if 1, or more, of the drives do not have enough free extents to create the LV. "2048 / PE Size / # of drives in each PVG" = # of free extents per drive needed. If each and every drive that you will use does not have that number, then the lvcreate will fail.

Do a 'vgdisplay -v /dev/vg05' and look at the end of the output for the # of free extents on each drive and if there is 1 with significantly less than the others, then you've found your problem.

[A. Clay Stephenson]

Your problem is that there are not enough PE's available on different PV's to grant your request. You can drop the -D y argument but you will lose your extent-based striping for this LVOL. Because you did not specify the PE-size, it almost certainly defaulted to 4MB but even the smallest (1MB) is really too big to see much benefit from the striping.

Drop the -D y and I think you will find that the performance hit that you think you will see will be very small - perhaps approaching zero.

[Doug_3]

Thanks to you all. I was under the false impression that each lvol would be round robin created and thus the pv's would be equally hit by each lvol. Striping in this scenario has absolutely no positive effect for me.

Could I have created the pv, vg, and lvol's in a way to spread IO across the disks? I have to use mirror-ux at the os level, but could have used the database mirrroring, which I think would have allowed me to stripe at a much smaller level than the 1 meg.

Thanks to you all!

[James R. Ferguson]

Hi (again) Doug:

You are correct insofar as the Distributed Allocation proceeds in a round-robin fashion. The "wrinkle" here is that you want to mirror too. The PVG-strict allocation policy ('-s g') insures that mirrors of the distirubuted extents don't overlap on the same physical volume. The easy (proper) way to do this is to use an LVM PVG file (man 'lvmpvg' for details).

Regards!

...JRF...