migrate from free esxi host to paid cluster with minimal downtime

[QuantumRiff]

Please excuse my ignorance, i am very new to VMware, and am familar with the concepts from other virtualization software I have used, but not some of the termininology, and perhaps that is why my searches are not turning up what I need.

We had been using the free esxi based hypervisor to host some server guests.  These servers run a copy of our database, and are used for reporting applications.  They are also setup with very large, thick provisioned VMDK files (>300GB!).  They have been running great for a year now.  However, we just purchased the VMWare "vSphere 4 Essentials Plus" kit for creating a 3 machine cluster, with some redunancy and other features that we wanted.  We also have a new san to go with it.  We are very happy with the product, and want to move everything to it!

On the other VM guests, running on this host, (and several other hosts, its amazing how they spring up) I would shut down the VM, and use the vCenter Converter to migrate the VM to the new cluster.  At the same time, I would convert from thick to thin provisioning. However, because these guests have very, very large thick provisioned HD's, its looking like they would take 5 hours or more to migrate. (other VM's have moved at essentially 1GB/min)  These reporting servers can go without updates from the master DB for several hours, but the servers still need to be available.. I can't have them down for 5 hours. There are lots of historical reports getting refreshed by end users.

Is it feasable for me to do the following somehow? I am having trouble figuring it out:

• disable DB updating
• shutdown the guest, take a quick snapshot, then start up the guest.
• While its running (and not updating) copy the main files, and VMDK file to the cluster (since file changes are going to the snapshot file)
• on the new cluster, once copied (or during the copy) convert the VMDK to a thin provisioned one (since only about 85GB of data is actually used on it). 
• lastly, bring up the new guest on the cluster at the same time I take down the guest on the old host (and just ignore the changes that have happened on the snapshot file, since its mainly a readonly report server)? and re-enable DB updating.

To the end users that would involve about 5 minutes total of downtime (basically, two reboots) which would be much easier to deal with than the 5 hours (each!) the way we have moved other VM's would take.

[AndreTheGiant]

Just use the serial code for ESXi and you host become a Essential Plus host.

To use HA and vMotion you must also build a vCenter Server.

Andre

[QuantumRiff]

Thanks for your reply...

I have tried adding it to the cluster (i have a vcenter server), but it doesn't seem happy to add it, since I have 2 machines setup with HA already.. perhaps that is because the two guests need to be powered down first? (the host's CPU hardware does not support the cluster's current Enhanced vMotion Compatibility mode.. the host CPU lacks features required by that mode)

But still, the essentials kit I purchased doesn't have storage Vmotion, so the guest would still need to be powered off to move the files to the SAN. (plus, we would need to add some fiber channel HBA's to that host to see the SAN)  Or am I wrong on my way of thinking here, or missing an important detail?

[DSTAVERT]

Could you leave everything as is and create a new DB slave? Let it catch up and then remove the old one?

[QuantumRiff]

that is a possibility, but a pain, as I would have to setup the slave DB, plus all the reporting tools, scripts, cron jobs. and of course, the exact same versions of everything..

[DSTAVERT]

Is it possible to clone without the data and then do whatever setup, rename, etc?

[QuantumRiff]

Okay.. wow, thanks for that tip!  I had just started looking into cloning my windows 7 guest systems.. (which so far seems much easier/faster than using MDT to redeploy images!!)  I had no idea you could clone a running system!  (I hope it does on Linux too!)  If this works, it will basically take care of steps 2-4 above!

So it would just be something along the lines of ?:

• disable updating the DB, and power off the VM's
• Add host to the VMWare cluster. (cloning doesn't seem to be an option in the free hypervisor)
• let it re-configure and setup HA, etc.. (maybe 10 min tops).
• boot up guest servers again on 'old' host as part of the cluster.
• Clone guest servers to one of the new hosts that is on the SAN.
• Power down guest servers on "old" host, and up servers on new host.
  • set new guest servers (the clones) to update DB (and ensure same IP addresses)
• Take down "old" host, and add HBA's for SAN..
• Enjoy Vmotion, and spread out my system load.
  • Man, I can't get over how handy vmotion is! its addictive!
  • evangalize the product

If that sounds right, I'll test it out on my linux servers this weekend!

thanks!

[DSTAVERT · Accepted solution]

If the standalone host has the same CPU type and family you will be able to add it to the cluster otherwise just add it to vCenter. You will still get access to the other features.

[QuantumRiff]

Thank you very much for pointing me to the right direction.  I am now cloning the live system.. and dancing a happy dance

[DSTAVERT]

The Snoopy dance or do you use something else? :smileysilly:

Good luck.

If you have time, keep this thread updated with your progress.

Thanks

[QuantumRiff]

Is there any other kind of dance?!

But in all seriousness, thank you again for the help!

for people that ever find this via google..

that worked great for my linux server (CentOS 4.7) and I'll try a CentOS 5.x one over the weekend.

also, the command

dd if=/dev/zero of=/empty_file.txt bs=1048576 count=100000; rm /empty_file.txt

is very handy to zero out unused space on a drive (thats about a 100GB empty file of zeros).  Makes the conversion to a thin provisioned drive a little faster and more compact.