Please excuse the brief TANGENT to illustrate why I love virtualization ...

Today a medium size company like ours might think it needs 200 servers.

But a quick look at the work performed by 2,000 people in my medium size company would show that for every number crunched, every document edited and every e-mail sent., there is very little actual computation performed.

Let's say each person in the company worked really, really hard and never took breaks. For all those numbers and documents and messages, they might need the resource equivalent of a million computer instructions every minute. Quick math says that the entire firm would need a capacity of 2 billion instructions per minute.

But how fast is a modern computer? Benchmarks of your basic $2,000 Extreme Dell PC show a capacity of several billion instructions per second.

So what gives? Why can't we run an entire firm on a Dell PC ... lot's of reasons, some due to the PC hardware design, most due to bad software design. But the truth is, if we fixed a few hardware and software design issues, many small to midsize firms could indeed run on the computing capacity of a Dell PC.

Bottom line ... we have lots more capacity on our systems than we need.

RETURNING TO MAIN ROAD from TANGENT

Virtualization makes it possible to exploit all that excess capacity.

But a fundamental flaw exists with virtualization implementations everywhere. Implementations are using far too many servers. Few are doing the math to determine their minimum computational requirement (MCR). MCR will tell you if you really need 40 servers or 4. Why is this important?

Think about the single system that can handle billions of instructions per second. What can it enable it to focus its tremendous computational power on business functions?

The POWER TO THE SERVER MANIFESTO or
How to Expose More Computational Power to the Needs of Your Employees and Customers

• Give processors fast direct connections to lots of memory
• Give processors fast and wide bus connections to the network
• Give processors fast and wide bus connections to storage
• Configure storage with lots of cache connected to lots of disks
• ...

If you virtualize on 4 servers, chances are you will be able to afford to invest in the Manifesto....
... and take full advantage of 40 years of Moore's Law

If you virtualize on 40 servers, chances are you will start making essentially bad buying/implementation decisions and try to use:

• servers with low memory capacity
• servers with processors that don't link directly to memory
• servers with narrow PCI buses and limited expansion capacity for the many network and storage adaptors required when hosting many virtual servers
• storage connected, not via dedicated high speed links, but via existing networks using iSCSI
• storage without sufficient disks (many per virtual server)

If you go the 40 server route, you will likely find:

• Your total computing capacity is not higher
• Your administrative costs are higher
• Your reliability is lower (more parts to fail)
• Your time between complete system upgrades is longer
• Your VMWare license cost is higher
• Your Storage and Network port costs are higher, and
• You won't be able to fully take advantage of all the latest features in your virtualization software including:

• • High availability
• • Live virtual server migration between physical servers
• • The ability to support high transactional systems such as Microsoft Exchange, Oracle, SQL Server 2007

A virtualization environment built with the limitations above will ensure you are forever managing a less than optimal infrastructure.

That was not our objective.

Part 2

Tags: X86 Mainframe, Virtualization