Does Virtualization Always Save You Money? (Hint: NO)
Eric Siebert’s post on “Saving money by using virtualization” has gotten quite a bit of screen time in my RSS reader this week as it’s passed around from blog to blog. I’m doing the same, however, I think his post may be more of a disservice to people thinking about virtualization. Don’t get me wrong: power conservation is a beautiful thing, and reducing the number of physical devices in a data center will absolutely save you money on raw power consumption as well as environmental/HVAC requirements.
But “going virtual” isn’t free, and unfortunately saving power isn’t the only metric used in data center ROI calculations. Adding virtualization to your data center can actually be extremely expensive. There are so many new considerations that have to be factored in when virtualization is introduced. Here are some cost considerations that Eric, and almost everyone else, glosses over so they can jump right to the huge, blinking $$$ savings:
- New Hardware Platforms (Virtual Hosts): These are typically more powerful boxes than the ones you’re replacing, have more CPUs and usually many, many additional drives and things like RAID controllers. These cost more up front than your single-purpose boxes and consume more power than individual, single-purpose boxes. It’s not a 1:1 power trade-off; a box that can run 50 OS images doesn’t consume 1/50th of the power that the physical farm consumed.
- Virtual Platform Licenses: This will obviously vary depending on which vendor you choose, but as an example, VMware ESX/i and Virtual Center ain’t cheap. Even if you’re looking at a less expensive option like Microsoft or Xen, you’re then looking at possible OS costs to run Xen, new expertise to manage it (see below), etc.
- OS and Application Licenses: The more guest images you add to a virtual host platform the less performance you’ll get out of each image (the law of diminishing returns). While you may have been able to load balance across 200 physical installs before, you might have to increase that install base to 250 or 300 once you virtualize because each one of those installs is able to run fewer processes, apps, etc.
- Management: You’re going to need to manage this new virtual data center, most likely in a different manner than you do with your physical servers. You now have to monitor performance and events of the virtual infrastructure and each individual hypervisor. So you’ll probably need to buy new management software as part of the migration, and your incident/event processes will probably change as well.
- Headcount: Saving headcount is almost always one of the top 3 drivers for virtualization: have less servers, have less people to manage them. And while that’s true, it’s probably also true that most of your existing IT staff aren’t virtualization experts, and have no idea how to troubleshoot ARP floods on a virtual switch, or how to performance tune multiple guest images on a host. You may actually have to hire new staff to manage your virtual infrastructure, or at least re-train your existing staff.
So again, I think that the core message behind Eric’s post is a good thing, but it’s missing the big picture. Thinking that saving on raw power is going to translate dollar-for-dollar into OpEx savings is short-sighted. Please do begin looking into power consumption as one of your data center cost metrics and as part of your overall virtualization strategy, but also factor in everything else that’s going to be required to complete this task. You may find that you save a ton of money within 12 months of converting, or you may find that savings is much less than you originally anticipated; just make sure you know that before hand and know what you’re getting into so you don’t promise your CIO $1M in savings only to spend $950k getting there.
April 29th, 2008 at 12:29 pm
The point of the article was focusing on one of the benefits of virtualization and not a complete analysis of the cost and benefits of a virtualization project. Obviously any enterprise virtualization project is going to cost money and your ROI will occur over time and will be a big factor in offsetting the costs of the project.
To give you a real world example of a project I ran:
Before: 80 physical Windows servers, 2 sys-admins
Project costs:
New Hardware Platforms - $90,000
Virtual Platform Licenses - $44,000
OS and Application Licenses - $0
Management - $0
Headcount - $0
After: 18 physical Windows server, 2 sys-admins
So for a one time investment of $134,000 we reduced our power footprint by 75%. To take that even further we would of had to replace at least 20 old servers anyway at a cost of $100,000 and we would of still had 80 physical servers with the same power consumption.
And to comment on some of your additional cost considerations:
New Hardware Platforms: In many cases you are replacing existing aging hardware to begin with, instead of buying new 1:1 replacements for your existing servers buy bigger servers that you can use to virtualize with. Replacing 12 smaller servers at a cost of $5,000 each with a single server for $25,000 just makes more sense.
Virtual Platform Licenses: Pretty much the price you pay to virtualize your servers. The virtualization market is becoming much more competitive and these costs will probably decrease as the non-VMware offerings become more polished. Competition is the best thing that has happened to virtualization, it keeps the dominant players on there toes and forces more aggressive pricing.
OS and Application Licenses: Most servers are under-utilized which is why they are great virtualization candidates. In most cases you should not need any additional licenses, if you do then you probably shouldn’t be virtualizing that server.
Management: For the most part you can use the same tools you used with your physical servers, you can also setup snmp alerts to forward to existing management software. Most management software vendors do offer virtualization specific modules these days.
Headcount: I’ve never really agreed with this one, typically virtualizing servers does not reduce the number of overall servers just the number of physical servers. So you still have the same number of servers to patch and maintain plus a few additional resulting from the host servers. Unless you have specific admins that only deal with hardware you will typically not reduce headcount.
If you add up all the costs versus all the cost savings over a 5 year period you will almost always end up saving money. Power savings is just one piece of that savings, energy prices are just going to keep increasing so power savings is a gift that keeps on giving. Obviously every environment is different and your mileage may vary but if virtualization was going to cost you money instead of saving you money there would not be many companies doing it.
April 29th, 2008 at 2:23 pm
Eric,
It sounds like you guys did a great job planning accordingly for your virtualization roll-out. But I think my overall concerns still hold true: almost every one I talk to who is planning a major virtualization roll-out includes cost savings as one of their top 3 reasons, and I think that overall that’s a red herring. It’s not the cost savings “smoking gun” that people who jump right into it are thinking it’s going to be. It’s like saying that buying a house will always result in an IRS refund. Yes, buying a house definitely helps your taxable income, and can result in a refund where you may not have gotten one before, but it by itself won’t guarantee Uncle Sam sends you a check every year. A few cases in counterpoint to yours above:
OS Licenses: You’re not buying new ones, but you’re not getting money back by not using them. And in fact, most customers will continue to use their existing licenses either in newly virtualized staging/dev environments or hold them in reserve for provisioning new images tomorrow. They may spin down 50 licenses at first but those are still valid licenses they’ve paid for. The other main point is that it’s so easy to burn through virtual OS licenses that they may be so excited they have 50 new licenses that before they know it they’ve repurposed all of them for non-production systems and have to go buy more in an emergency.
Management: Sure, you can manage your guests w/o any change, and plugins do exist for basic hypervisor status info, but not for full-scale provisioning. Implementing VMware’s Virtual Center and VMotion costs money. You don’t need it, but most people go with it (or something of the like). If you’re building a complete virtual data center, you’ll need a management solution that knows how to manage the virtual components, not just a plugin.
So again, I love it that people are able to save money by implementing virtualization, when they plan accordingly, and I do think it should absolutely be part of the overall consolidation/green/efficiency planning. My biggest beef is that people all too often make a 1:1 relationship between virtualization and free money, and that just ain’t the case.
Thanks for the discussion…
-Alan
May 12th, 2008 at 4:51 pm
[…] focused on the savings created by decreased data center power consumption, his blog received this response on the Virtual Data Center blog: I think that the core message behind Eric’s post is a good thing, but it’s missing the big […]
June 15th, 2008 at 7:04 pm
virtual image
Shinya focused on techniques that would allow for successful operation of the endoscope by an individual, rejecting the common practice