Virtualization: Spin or Wobble? (Part 1 of 3)
When you look closer, the spin around VMware benefits starts to wobble quickly when you do the math.
Hardly a day goes by that my e-mail does not contain a missive from another vendor announcing support in its product for VMware. You would think that the EMC majority-owned server virtualization software player was so important to the IT universe that all other technologies revolved around it.
The company's marketecture certainly reinforces that message. They argue that smart companies have a "VMware strategy," which is kind of a weird idea in and of itself. VMware, at best, is a set of software tools. It is not a strategy unless you regard a wrench as a strategy.
Server consolidation is a strategy, and IT resource optimization might be a strategic objective, but calling VMware a strategy is a bit of an oxymoron.
Analyzing the Analysis
When I read analyses on server virtualization, and there is no shortage of them, I am impressed by two things. First, for all the sound and fury, the numbers just don't seem compelling. At last count, one of the prominent analytical organizations was claiming that sufficient licenses for server virtualization software were being sold to virtualize about 30 percent of the servers deployed in the world. I suppose that sounds significant, but of that pie, about 37 percent of the servers were equipped with VMware licenses, with other slivers going to other vendors of hypervisors -- small providers such as IBM, Microsoft, Oracle, Citrix, Virtual Iron, etc. Based on this, VMware is just another big fish in a small pond.
In fact, the small pond becomes more of a small watering hole when you realize that IDC and others are counting all licenses shipped with all servers. Since Dell and HP are bundling basic VMware with every server they sell, and since Microsoft adds Hyper-V to every OS they sell for a paltry sum ($18 last time I checked), this number looks huge. However, no one is saying how many of these licenses are actually used.
Better information might surface in the coming year, when the stock analysts watch VMware's license renewal numbers. According to one significant stock watcher, the first slice of enterprise account renewals in early 2010 will tell the tale of whether their new rating of VMware stock ("underperform") was correct. They cite increasing reluctance of enterprise accounts to buy anything, combined with a slowdown in server sales with bundled virtualization wares, plus the continuing problems of OEM revenues in arrears, as justifications for a negative outlook on license-based revenue growth for the company throughout the balance of 2009.
Indeed, the spin around VMware starts to wobble quickly when you do the math. There is no such thing as a VMware strategy, only a lot of follow-on work to get the technology to perform like what is represented in the brochure.
According to analysts, the most popular use of server virtualization is to consolidate file servers and low-traffic Web servers, neither of which requires virtualization for consolidation. I run about 60 Web sites under a management tool called Plesk and eschew virtualization because the Web sites simply don't require separate virtual machines and may even become performance and management impaired if I deploy them in VMs.
When I consolidate file servers, which I do from time to time as part of a normal hardware refresh, I use that old-school virtualization technology called the file system. I simply take the contents of the existing file server that I am about to retire and copy them into file folders on my newer, higher-performing, consolidated file server. If you look around my infrastructure, you might well find a folder named "Old Pikachu" or "Old Snorlax" -- naming conventions that recall the days when I named all of my servers after Pokemon characters.
Truth be told, I probably could have migrated all of the data from old file servers into an archive comprised of green media, such as tape or optical disk. Last year, at Usenix, researchers released findings of a survey of about 80,000 NetApp network-attached storage Filers revealing that most files had not been touched in over a year. I remember thinking that my next NAS should probably be a tape drive given the characteristic re-reference rates of user files, which tend to decline exponentially after just a few days.
Not Without Value
For all the hype, server virtualization isn't a strategy, nor even a compelling trend to me. I do see its value as a potentially useful tool, however, in two areas.
First, application test and development operations have often used server virtualization to supplement their limited server resources. I understand that. When performing software tests, it is not uncommon to group several virtual machines to create targets or stress-testing sources. You tear these virtual servers back down when the work is done. I understand the utility value of server virtualization in this context.
Second, disaster recovery can sometimes be facilitated by server virtualization, though not in the manner described by VMware. I worry that stacking several applications in VMs on Tinker Toy x86 server boxes is actually a disaster multiplier. This is not ameliorated by the high-availability (HA) failover capabilities touted by VMware and a few other v-server vendors, because in our test labs, their promised failover capabilities don't work most of the time.
What I do see as a potential play for server virtualization in the data recovery space has more to do with lowering cost in the recovery environment. I have helped many companies economize on their recovery plans by using minimum equipment configurations for restoring critical systems. Virtualization of servers may play a significant role in building minimum equipment configurations.
As the University of Texas at Brownsville (not my client) demonstrated last year, it is entirely possible to recover a highly configured Microsoft Exchange environment (some 40K mailboxes hosted on clustered servers connected to a back-end Fibre Channel fabric) on a virtual machine located in a 1u rack-mount server with direct attached storage within an Austin, TX ISP -- and to operate from that platform for a short time. Given the much-reduced workload typically run on systems operating in a recovery environment, server virtualization might be harnessed to reduce the costs of such a DR infrastructure.
In production environments, however, I am wary of the whole VM pitch. Some reasons have already been explored in highly technical publications such as the Wall Street Journal. Not long ago, they noted that companies adopting VMware and other virtualization wares in order to contain server sprawl were now struggling with "virtual server sprawl." No one can readily discover which physical server is currently being used to host which virtualized application when a hypervisor is moving VMs around infrastructure at will.
In fact, hypervisors are a topic all to themselves, which will be my focus in my next column.
Your feedback is welcome: firstname.lastname@example.org