VMWare Extents 4.1

[jdelpero]

Hi all,

This topic is endlessley debated...

I have a 3 Host setup connected to a single DS3512 IBM SAN.

the SAN has a single RAID6 Volume which I have carved into 2 x 2TB Luns and 1 x 1TB LUN. (The SAN is fibre attached with redundant fibre switches)

I am only running 6-7 VM's a few of which require 1TB HDD.

My Question:

Why is it not ok to use extents to create a single 5TB datastore...? It's the same SAN, the same RAID array and connected by the same fibre cables.

All I can see out there is no no do not use extents. If a RAID controller dies in the SAN the other takes over - ie. If I lose one LUN I will only ever lose all LUN's so why have I been told that it is still to dangerous to use Extents??

Thanks for your input.

[idle-jam]

yes you lose one you lose all. also the MAX vmdk is 2TB or less hence going expand does not make sense as you will still have multiple VMDKs.

[jdelpero]

See thats what I don't really get...

Because its the same san and raid volume even if it was configured without extents (seperate datastores) i would lose them all at once. There is no way for one LUN to stay alive and another not to??

If the RAID6 Fails all Lun's drop.

If the RAID controller fails the redundant RAID controller kicks in and all  LUNS stay alive.

How could one LUN drop and another stay alive it's just not possible?? Also I have configured SAN mirroring and FlashCopy. Veeam is replicating twice daily so from a backup perspective it's safe.

Can you shed some light on why extent is so evil in this situation?

Have SAN's been known to corrupt a LUN ie. specific datastore corrupt thus in an extent situation would destroy the entire datastore? but without extents the other datastores stay alive...?

The reason why I want a single Datastore - Maximise use of datastores (lots of small datastores are very inefficient as you can't fill the space with thin disks correctly. Snapshotting space. ie. If i have two vm's with close to the 2TB provisioned I run out of snapshot space. yet another datastore may have plenty of space as the vm's aren't using all there thin provisioned space. (thus easy for client to manage storage space)

[idle-jam]

it's rare to have a single LUN failure, but just in case if someone just screw up the MBR and etc. that's a if. but in your case if already have veeam then it's safe to go ahead with a single datastore ..

[depping]

You need to ask yourself why you would do it?

1) max file size is still 2TB

2) larger volumes means more stress on a single volume (think scsi reservations)

3) VMFS volumes with extents are more difficult to repair when it comes to the VMFS filesystem

4) If your array has VAAI capabilities than it will not leverage these with extents

to be honest, I don't see many valid arguments to go that route.

Duncan

HA/DRS technical deepdive - the ebook!