Get started Bring yourself up to speed with our introductory content.

How to determine which physical servers are good virtual machine candidates

Having trouble determining which physical servers should be virtualized? Expert Ron Oglesby offers his thoughts in this expert response.

We are having a debate in my organization as to the use of VMware ESX. One side of the house is looking to limit it to the lab as they feel it doesn't perform as well as real hardware. My side of the argument is that most of our servers don't use the physical hardware to its limits so VMware would be fine. Is there a way to determine which servers would be fine as virtual machines (VMs) and which wouldn't?

Great question (and potential five page answer). I see this all the time, and the only way to answer the question is with math and real numbers. It is impossible to say any server will be 'fine' as VM. Sure it will run, but will it perform its tasks as it does today without the end user noticing (and that is the key, does the end user see a difference?).

Of course most people don't want to sit next to some users after a migration and find out, but instead want a way to determine candidate potential ahead time. We do this (and have done it thousands of times) by rationalizing what is used today to real numbers.

Let's take processor usage: If an existing machine has a dual processor 3 Hz server (total of 6GHz), but during the day only averages about 10% utilization, then the machines needs 600 Mhz. Just fine for a single processor VM. We have even ranked tier 1 candidates up into the 1.5GHz range.

If the application is running takes advantage of SMP, then maybe a dual processor VM would work for it. From a memory perspective, it is pretty much a wash. If your machine uses 1.5GB now, it will use 1.5GB as VM. Of course, ESX-based VMs can now go at 16GB of memory, but who would want to at today's memory costs for 8GB DIMMs.

For disk and network there is a certain penalty to pay. On the disk side it is about 10-20%. But of course most Wintel servers never average above 10 or 20 IOps a sec, or ever get near the SCSI card's throughput capabilities.

On the network side the penalty is a bit higher. Then again, a recent client I worked with who had 1000+ servers being surveyed only had about 50 or so that used more than 1MB on average… does every server really need dual redundant 1GB NICs all to themselves?

Show your coworkers the numbers. Work from the facts and your argument will be more persuasive.

This was last published in July 2007

Dig Deeper on Reducing IT costs with server virtualization

Have a question for an expert?

Please add a title for your question

Get answers from a TechTarget expert on whatever's puzzling you.

You will be able to add details on the next page.

Start the conversation

Send me notifications when other members comment.

Please create a username to comment.