Point of no return

Well, we have reached it and passed it. The production servers have been ordered as of 5 minutes ago.
Although PowerConvert will probably only help with the Windows servers we have enough confidence in VI3 Enterprise to go ahead with the migration of our production data centers.

Platespin issue resolved…. and new one gained

I’m glad to report that the Platespin issue we had wasn’t a license issue but an error on our part. Although the documentation that Platespin provides is a little sparse in certain areas, their customer support talked us through and we found out what we were doing wrong.

What we did was to try to deploy a P2I image of a server to the Virtual Center Management server instead of deploying it to one of the VM host cluster members. Second problem was that the user that we used to discover the ESX server with did not have the correct access rights to complete that process successfully (because you can’t use root for that). You need a user that is allowed to log on through ssh and has the correct rights to the management console.

After resolving this problem deploying the image worked with the addition that Smartstart deployed servers (in this case: Proliant DL360) have a lot of HP/Compaq device drivers that load on startup (as service). As that hardware no longer exists in the VM you get the Windows error message about services that failed to load on startup.

Solution was to remove these programs after conversion as we don’t need them anymore.

This concluded the test conversion of a bog standard Windows box with SQL server on it. Having gone through all the steps the test conversion of a SLES9 box went smoother, at least for the first part.

This particular server (and 90% of our other Linux boxes) has LVM on it to manage the diskstorage and PowerConvert seems to be incompatibel with that. First try delivered a kernel panic to suggest that. After removing the LVM partition and making it a reiserfs volume conversion worked to the point that the new VM server saw it’s own storage and could eventually boot.

We don’t know yet if and when LVM support will be added to PowerConvert so until that time only the Windows servers can be PowerConverted. Which is unfortunate as roughly 50 servers are Linux (50% of the total). A call has been placed with Platespin and our distributor to find out about the LVM support.
We’ve been thinking about a workaround for the Linux servers and so far can only think of just deploying a bunch of new VM’s based on our working SLES9 image and then copying the config files and data over. It’s not really creative and has the potential of becoming a lot of manual work so we’ll keep on looking for alternatives.

The joys of project planning

I always get a little edgy when people state that “project plans are there to be changed” but it seems we’ll be using that quote quite often this period.

Because we still have the license problem with Platespin we made zero progress on that side and there has also not been a decision about the storage environment.

As I will be off for a week starting next Friday so I am hoping Platespin can resolve the issue and get us a working Proof of Concept license tomorrow. If all goes well (we have made successful P2I conversions so the only thing standing in the way of a successful test is the I2V deployment) then the orders can go out to Platespin and HP for the hardware and software before I leave.

Planning now looks like this:

• Ordering production VM hosts: 31-8-06
• Decision on storage: hopefully 1-9-06
• VM host cluster online: 18-9-06
• Production tests: 18-9-06 to 1-10-06
• First storage array online: 1-10-06
• Writing migration plan: from 7-8-06 to 18-9-06
• Start of migration: 1-10-06

Workstation vs. Server

I was a bit slow to find this document on Virtualization.info but found it very useful. It’s a comparison between VMWare Workstation and the newly released VMWare Server.

If you’re wondering what product to use for which task it’s a nice compact overview.

Barracuda rocks!

Completely unrelated to virtualization but I just have to get this out:

The Barracuda Networks Spam Firewall rocks ! 

We have done a trial of a week and cought 12% more spam than we did with our old solution (SpamAssassin + MailScanner + MailWatch) which required more and more manual configuration and tweaking of detection rules. Although the Barracuda is still partly a SpamAssassin implementation they have reworked the Bayesian system among other things.

We went to full production last Tuesday and with some more training of the Bayesian we are now at a whopping 18% increase in spamdetection.

The units have performed flawlessly so far (in active-active cluster) and we got to ditch our old SMTP servers as it does outgoing and relaying as well. With the new firmware it even does outgoing spam checking. It just sweet. And price-performance is a no brainer as we can depreciate them over three year.

PowerConvert issue + storage status

PowerConvert has been giving us some headaches with the licensing. We have a Proof of Concept package but the license doesn’t allow us to deploy it. A call has been entered with Platespin and we are awaiting their information.

Also the decision on storage has not been taken yet because of political issues. EMC has our preference but there are other factors at play.

The final configuration looks like this:

• Clariion CX3-40 with:
  • 2 cages with 146GB 10k FC disks (for transaction systems and VM’s)
  • 1 cage with 300GB 10k FC disks (mostly for fileserver data)
  • 1 cage with 500GB LC FC disks (LC = LowCost / for archiving of data and diskbased backup)
• 2 Cisco MDS 9020 SAN Switches
• EMC DiskXtender
  • This is software that does policy based file server archiving (if it goes to EMC we’ll be adding EmailXtender in 2007 for policy based e-mail archiving)
• SanCopy
  • This is software to make array based copies between different SAN’s.
• Snapview
  • This is software to make snapshots and clones
• Navisphere
  • SAN Management software

Archiving for now will happen straight to the 500GB disks. In the future and taking into account changes that could happen in the field of data retention and new legislation we have the option of adding more near-line storage or adding Centera’s.

We will keep our MSL5026 robot for now and do backup and e-mailarchiving in 2007. We wouldn’t have the personnel for it anyway and it saves some serious money in the budget.

I consider myself lucky that we do not have to be SOX compliant or overly apprehensive because of liability. There is some privacy law that we have to live with but the consequence for the infrastructure is minimal. For example: BCM and our ISO27001:2005 certification are more important at the moment.

Have good weekend,

Looking forward

Once we finish testing PowerConvert and some last loadtests we will start with setting up our production environment. The definitive decision on the new SAN has not been taken yet so that could still become exciting should we have some unexpected external influence.

EMC has our preference but there could be some political influence in the direction of IBM. We’re hoping to strike the right deal financially with EMC so the discussion can focus on functionality again.

The purchase of the storage environment has now been classified as being “on the critical path” because it will make or break our planning. Our escape route is to buy yet another disk cage to attach to one of the MSA1000’s we currently have but that would be a waste of the money in the long term and we would only be able to run production on that for a very short time (no redundancy).

Our milestones for the coming period are:

• Ordering production VM hosts: 15-8-06
• Decision on storage: between now and 25-8-06
• First storage array online: between 12-9-06 and 18-9-06
• Writing migration plan: from 7-8-06 to Fri 8-9-06
• Start of migration: Mon 18-9-06

What not to virtualize

Just a quick update to let you know that we will start testing PowerConvert this week. Tests will probably continue into next week but things are a bit slow due to the holidays at the moment. If all goes well with PowerConvert I will start ordering the first two production environment machines from HP.

We have also made a list of the things we will not be migrating:

• The Faxserver
• The Exchange Servers (not until full 64 bit host support so we can upgrade to Exchange 2007)
• The Cisco Callmanager servers

These are the things we know for sure. If anything changes I will post it here ofcourse.