As I continue to talk with small businesses about virtualization projects, several themes have come to light. The company has tried virtualizing but it never worked right, or it was to expensive, or we didn’t know enough. These issues draw a line back to one problem. The company doesn’t understand the resources required to create a virtualized environment.
Let me start with a typical example. In most companies an engineer in the I.T. department has hoarded some hardware and has some type of hypervisor installed on it. Things are working O.K. but either few or no servers have been used in production. The company has some knowledge and wants to move forward with virtualization, the system engineers are all for it because it will make their lives easier, management has read over the ROI reports and get on board with the hype. Now the rubber has to hit the road, the first big purchase of hardware. Up until now the company has been buying servers one at a time, only putting out $10k to $15k a piece. Now we get an order put together for an HP C7000 blade centeer with fiber channel NetAPP storage or similar configuration. A simple initial order like this can run up to $300,000. Back to the drawing board. Another configuration is devised that uses an HP C3000 and 10g iSCSI storage. This plan is then presented to the powers that be. The jaws don’t drop as fast at the new $150,000 price tag. But still that price is a no – go. Now you opt for four 1U servers, 1G x 4 port iSCSI storage, some dedicated iSCSI network cards, cut the disk space in half. Finally the company swallows the $70,ooo.
Now comes the problem, instead of a system highly designed and targeted toward virtualization, you are given a system that you have to somehow make work. Instead of 8g Fiber Channel, or 10g iSCSI, you have basically 4gig of trunked iSCSI storage. The class of storage is now less configurable and less flexible than a NetAPP, EVA or EMC. The servers are similar buy now instead of being able to manage them from a blade center console you have to update them separately. Soon after you begin loading virtual machines for testing you realize that with less storage you now have fewer spindles to run the SQL server and it slows down. Other virtual machines are effected because instead of multi-path fiber you are using single path iSCSI. You begin the get the point that unless the company is willing to put a complete investment into virtualization it can be doomed to fail. As an Sr. Systems Engineer it is our job to education management on what it will take to have a successful virtualization deployment.