Do I need to delete the partition and recreate it in linux before I re-create the storage repository (or is it ok if the underlying partition has LVM setup)

THANKS FOR THE HELP!

Very helpful article on setting up thin provisioning with ext3. Just what I needed.

Regarding the template “XenServer Transfer VM 5.6.0-31124p” that needs to be uninstalled.. It’s actually pretty important. This is what handles all the functions for the Virtual Appliance Tools (Import Virtual Appliance, Export Virtual Appliance and Disk Image Import). These will all fail without this template.

Fortunately it is VERY easy to place it back on your server without the need for exporting from another XenServer and importing it…

In the XenServer host directory /opt/xensource/packages/files/transfer-vm/ are a number of related files. It takes a couple easy steps to reinstall the Template VM:

1. If you CD into the directory and list the files you will see the output::

[root@ixen ~]# cd /opt/xensource/packages/files/transfer-vm/
[root@ixen transfer-vm]# ls -al
total 3692
drwxr-xr-x 2 root root 4096 May 12 12:30 .
drwxr-xr-x 5 root root 4096 May 12 12:30 ..
-rw-r–r– 1 root root 134 May 20 2010 65-install-transfer-vm
-rwxr-xr-x 1 root root 478 May 20 2010 do-copy
-rwxr-xr-x 1 root root 613 May 20 2010 do-transfer
-rwxr-xr-x 1 root root 1784 May 20 2010 install-transfer-vm.sh
-rw-r–r– 1 root root 3747840 May 20 2010 transfer-vm.xva
-rwxr-xr-x 1 root root 491 May 20 2010 uninstall-transfer-vm.sh
[root@ixen transfer-vm]#

2. Then simply run the install-transfer-vm.sh script and you are done:

[root@ixen transfer-vm]# ./install-transfer-vm.sh

Hope that helps some people.

Thanks for your article.

http://www.scriptkiddie.org/blog/2010/09/19/reducing-server-io-on-virtualized-hosts/

If you’ve got ubuntu, you’ll probably need to adapt the instructions.

By compression I mean physical-to-virtual compression — 10 images on one virtual server being 10:1 (VM) compression.

Part of the issue that I’m dealing with is having literally hundreds of dell 1950s chassis that we already own and having zero budget for capital.

Interesting to know that you’re getting away with using internal storage with 35:1 VM compression (although you’ve got more drives). I’ve got some old 2950s with 6 drives that I’m looking at using and a few DL380s with 8 drives that are in the same class.

I haven’t tested out the ext3 I/O, I’d be interested in any numbers you come up with if you test it.

I’ve also been thinking about blogging something about tuning to reduce I/O on servers.

Can you tell me what 1U servers you’re using? Also I’m a bit baffled by your ratio. You said you’re not using shared storage but you mentioned compression? How are you getting the compression? Or were you talking about thin?

We’re thinking of using nexentastor for shared storage although thats still a bit further away (and the deduplication is currently buggy). I’m not sure how reliable it would be but it shows promise.

Currently we have about 35 VMs (mix matched centOS and server 2008) per host (8 cores and 64gigs of ram, 3ware raid10 8 drives) and it works pretty well. But it is using the default install and not ext3 which is why I was wondering how much of the impact it would be if I did change to ext3.

Thanks for your tips on the I/O. Will check it out.

At that point you probably want some kind of SAN/NAS appliance that does disk de-deduplication natively for you.

Right now I’m taking already purchased 1U servers and chopping them up into more like 10-16 VMs with XenServer Free. One thing that I do is eliminate all the useless cronjobs that chew up I/O in the middle of the night (slocate, etc). I’ve also turned off fsync() in syslog.conf which produces large wins in getting I/O down very low.

Since there’s no licensing overhead, or shared storage or anything I’m just getting 10:1 or 16:1 compression and making efficient financial use of capital (each server cost $5k or so, 3+ years ago). If you’re buying expensive virt heads for $20k and loading them up with Enterprise licensing, then you need to hit compression like 50:1 to make sense, and then doing ‘thin provisioning’ this way doesn’t make any sense.

You don’t lose vm snapshots if you do this, but they’re not as fast or thin. I can still take vm snapshots and then vm export a file to take a backup. Tends to crush the I/O in the middle of the night, though.

And I haven’t measured what the relative I/O performance is between ext3 and LVM or LVHD, but like I said, I’ve tuned the VMs that I’ve configured this way so that they mostly don’t do any I/O…

And, I don’t recommend this for production or load testing. I’m using this for cheap virts for corporate IT stuff, and integration and dev sandboxes. It is very “poor-mans” approach.

Also I assume you also lose the LVM snapshot ability which is pretty handy.

So far I don’t have any shared storage to play with. Right now I’ve only got corp IT (wikis, etc) that I’m reusing some old crappy hardware on. VM migration is somewhat painful with vm-export/vm-import to NFS for backups and when I have to move things around. So everything in my environment right now is a little bit ghetto — but this trick is helping me keep the disk space down.

I’m thinking of putting together some cheap Linux iSCSI + GNBD-replicated shared storage heads for pre-production use.

And at some point I’ll need proper shared storage for load and production.

For both of those, I’m just not there yet, though, and it’ll be a few months before I get there.

Is that even possible?

If so enlighten me :)

Thank you in advance