Servers

IT cost containment through virtualization: One way to do more with less


Over the next few weeks, some of my posts will revolve around containing or reducing the costs necessary to run your IT operations. I'll try to present as many angles as possible. Further, the items I present are things that we're actually implementing--or planning to implement--in my organization.

By now, you've probably read a million articles extolling the virtues of virtualization... and here's one more, but with a twist. Believe it or not, virtualization isn't a financial panacea, although, done right, you will probably enjoy some savings. Or, at the very least, you might be able to improve efficiency and availability while saving money or without much in the way of additional cost. Rather than simply spout well-known facts back to you, let's use a scenario instead. Before I start, please understand that this is a very rough scenario. It's not going to take into consideration every possible factor, but will provide a general indication of whether or not virtualization is a good idea for an organization. For our example organization, let's use the following assumptions:

  • 30 typical x86 servers with an average replacement cost of $7,000 each (reference: http://www.news.com/Server-buyers-shift-toward-muscle-machines/2100-1010_3-6169126.html) - this money is already budgeted.
  • Existing servers are dual processor, dual core machines.
  • 22 of the servers run a typical x86 workload. For this example, let's assume that each server runs at 10% utilized. Of course, there will be occasional spikes.
  • 8 servers are higher-transaction, more-highly-utilized servers - say 40% on average. Assume that these are SQL or Exchange servers.
  • A 3-year server replacement cycle. 10 servers are replaced each year.
  • Around 10 TB of total disk space is being utilized by all of the servers. The data growth rate is about 25% per year. The higher transaction servers currently hold only 2TB of the 10TB, but have the same growth rate.
  • The organization is using all direct-attached storage, but wants to increase overall service availability.

Now, let's also assume that this is a pretty cost-conscious organization that doesn't have a huge budget. As such, they don't have a ton of money the throw around on new hardware and software and they need to make the most of their budget. This organization has the following goals:

  • Higher availability.
  • Reduced or similar costs.

One possible solution

There are a million different ways that this fictional organization might go about achieving this goal, but we'll consider this possible solution. Currently, the organization has budgeted 10 server replacements per year for a total $70,000 per year. In order to achieve the goals set forth, let's use the following solutions:

  • New servers will be dual processor, quad core machines. All but 6 existing servers will be virtualized to VMware ESX. 19 of the 22 typical-load servers will be virtualized and 5 of the 8 higher-utilization servers will be virtualized. Virtualization will be phased in as servers come due for replacement. For typical load servers, we'll run about 6 VMs per ESX host. For the higher-load servers, we'll assume that 2 or 3 VMS will run on each ESX host. Although new servers won't need much storage since they'll be connected to a SAN, they will need lots of RAM, so we'll keep the same average server cost of $7,000.
  • VMware ESX Enterprise 3.5 (Virtual Infrastructure Enterprise 3.5) will be used on all ESX hosts. ESX Enterprise 3.5 carries a list price of $5,750 for a two processor license. We'll use this list price in this example. Enterprise has all of the availability features necessary.
  • Storage will be consolidated to a SAN. For this example, let's use EMC's new AX4 as the storage array of choice. Why? I've been looking at this pretty hard in my own organization lately and I'm most familiar with this solution. The new storage will use both SAS and SATA disks, depending on the need. Server storage will also be migrated to the SAN as servers come due for replacement. Assume that each array costs about $25,000. This is a little less than list price, but reasonable and includes 3 years of maintenance.
  • Storage arrays will use RAID 5. Usable SAS-based array capacity would top out at around 4TB while SATA-based array capacity would top out at around 10TB (1TB disks are due to be introduced to the AX4 in March).
  • We're using nice round numbers for this example. In reality, your analysis might not be quite as straightforward.

Let's start with the storage side and look at the three year storage needs:

 

Requirements

  SAS SATA
Year 1 2 TB 8 TB
Year 2 2.5 TB 10 TB
Year 3 3.125 TB 12.5 TB

As far as storage goes, in Year 1, we'll also be buying two storage arrays. Each array costs $25,000 for a total of $50,000. We'll also buy two ESX host servers at $7,000 each and two ESX Enterprise dual-processor licenses for $5,750 each. Further, in year 1, we'll replace two of the physical servers that will not be virtualized. So, year 1 costs will total $89,500. One of the ESX hosts will house the workloads for six existing x86 virtual machines. The second ESX host will house the workloads from two of the more highly utilized servers.

In year 2, we'll add another SATA-based array for $25,000, will replace four servers (2 physical machines and 2 ESX hosts) and will buy two dual processor ESX licenses. Year 2's costs will come to $64,500. Six more typical physical machine workloads will be moved to one of the new ESX hosts and two higher-capacity workloads will be moved to the other ESX host.

In year 3, no more storage will be required since there will be adequate capacity on the existing arrays. However, four more servers will be purchased at $7,000 each. Again, two more dual processor ESX licenses will be purchased at $5,750 each. Total outlay: $39,500.

So, at the end of three years, we're looking at a total outlay of about $193,500, or about $16,500 less than what would have been spent if servers were directly replaced. Again, the numbers above are very rough, but give you an idea of what you're looking at. Beyond the direct savings, even if it's somewhat minimal, you've achieved the following goals:

  • Implemented a more robust infrastructure. With VI3/ESX Enterprise, you get VMware's powerful high availability features, including VMotion.
  • Developed a responsive architecture. Now, when you need to deploy a new service, you don't need to purchase, deploy and provision a physical server. You simply create a new virtual machine on the appropriate ESX host.
  • Implemented the beginnings of what could be a disaster-resistant solution, depending on how you physical deploy the hardware. Again, this would only be a beginning, but you have to start somewhere.
  • Done all of the above without increasing your costs. Sure, year 1 costs are a little higher, but this increase is more than eliminated in year's 2 and 3. And, if we went to year's four and five, you would see even more dramatic savings since the deployed storage arrays still have enough capacity even given a 25% annual increase in storage needs. If year 1 costs absolutely must stay within the annual $70,000 budget figure, you have a couple of options. The first would be to run some of your existing servers for another year and spread the year 1 costs out a little. The second option could be to put all of your equipment on a multiyear lease.

Summary

The example I provided could provide even better results if more VMs were run on each ESX host. I was very conservative in this example. If you're using dual quad-core systems with a lot of RAM, six VMs per host is paltry. You could likely double that number. However, I didn't include some ancillary costs that you may want to include, such as maintenance on your ESX licenses and employee training.

I hope this post gives you another way to think about cost containment/infrastructure improvement. I'd love to hear some of your ideas on this topic. Some of your ideas may make their way into future posts, too.

About

Since 1994, Scott Lowe has been providing technology solutions to a variety of organizations. After spending 10 years in multiple CIO roles, Scott is now an independent consultant, blogger, author, owner of The 1610 Group, and a Senior IT Executive w...

2 comments
Tony Stanford-Beale
Tony Stanford-Beale

Couple of comments: Good for the uninitiated but there are some other good reasons/ things to consider as well: 1. Power consumption and cooling. Hopefully by reducing/ replacing the physical number of servers you actually reduce the per processor power consumption - its worth checking! less power in means less heat out therefore less power to cool! Lots of $ savings! Whilst the actual power per box may be higher it is being more efficiently utilised... and moving direct attached to SAN improves that equation as well. Do check the cooling arrangements when you plan to virtualise especially if going to blades. 2. Labour costs. True you still operate the same number of logical machines but overall effort to manage should improve and time to perform a task eeg a restore or reconfigure is more efficient so consider the hidden costs as well. 3. For those who are already virtualising servers, there is plenty of other things to virtualise - networks, storage, etc.

Michael Kassner
Michael Kassner

Very informative read, especially with using a scenario as an example.

Editor's Picks