In-Depth
Virtual Servers and Storage: Truth or Dare
How many shops are really using server virtualization?
Usually, I don't heed polls and surveys unless they are "scientific" -- that is, conducted using a proper sample size and demographic characterization, and non-leading questions. Even then, I worry about the "PBS Effect" that I learned about in political science class many years ago: everyone wants to appear smart, so they say that they watch intellectually stimulating programs on PBS when, in fact, they really watch Jerry Springer, Survivor, or other popular fare guaranteed to cost viewers IQ points with every passing minute.
That's one reason why I hesitated to interview Symantec's director of storage and availability management, Sean Derrington, about the results of a survey commissioned by his company and executed during the just-concluded Storage Networking World in Orlando, FL. The poll included 100 non-vendor attendees at the show, predominately storage managers and administrators or IT managers, 64 percent of whom were from larger firms and 19 percent of whom were from smaller companies.
Was this sample size scientific? Could the viewpoints of only 100 folks really provide insights into the complexities of server virtualization and its corresponding storage-related woes? Derrington's argument was simple: "One hundred questionnaires are better than 50" -- and an improvement over the case-based reporting he was seeing in the marketplace of ideas.
So what did the survey reveal?
First, everyone was using server virtualization. This finding alone contrasted starkly with data from large industry analyst firms that placed server virtualization strategies in only about 25 percent of companies. That this study found the preponderance of respondents using VMware as their virtualization technology was also somewhat inconsistent with analysts’ numbers. Industry analysts say that of the 25 percent using X86-based virtual machines, only about 38 percent use VMware. It is your classic “big fish in a small pond” story.
The last key finding of Symantec’s survey was that 73 percent of the respondents said that they had already placed server virtualization into production (or planned to in the short term), while most industry analyst reporting has it the other way around. According to most analysts, about 75 percent or so of those firms that are experimenting with virtual machines are doing so in test and development environments and have not placed VMs into production platforms. The Symantec study strongly contradicts this number.
What could account for these discrepancies? I haven't a clue, except that attendees at this year's SNW may have been there in large part because of storage issues they are having with their X86 virtualization environments. This alone would have provided a stilted sample.
Another possibility is that the analysts could all be wrong. Gartner once claimed that we would replace our PCs with Network Computers, and IDC told us at about the same time (in the late 1990s) that we should stop buying shrink-wrapped software because all software would be provided via application service providers across the Internet. Now, Gartner claims that Microsoft's Armageddon Day is upon them, while IDC is claiming that storage capacity growth isn't keeping pace with data growth and that this discrepancy will soon produce a genuine storage-gap crisis.
On the other hand, given the hype around virtualization today, saying that you don't have a server virtualization strategy might be perceived as a social negative. Like cigarette smoking or driving a non-battery-powered-automobile, the lack of a server virtualization play might make you appear to be less hip and less cool in some circles.
Bottom line: Symantec has some numbers and they are sharing them with every reporter who will sit for a thirty minute overview, mainly I suspect, because of the list of storage woes that the survey generated. It seems that there are significant storage pain points among the X86 virtualization group. Here's the survey's breakdown.
Forty-nine percent of respondents claimed that storage management was one of the "top challenges in VMware environments;" 45 percent claimed that backup/recovery was high on their list. Twenty-nine percent of respondents indicated their data protection software for virtual environments was "lacking" and 20 percent said that they resort to backing up entire physical server environments instead of discrete virtual machines and their data, which is apparently a more daunting task. Only 30 percent of respondents said that recovering individual files from a virtual machine was "easy."
Other interesting results showed that two-thirds of respondents back up at least 60 percent of their application data to tape. Thirty-three percent said that they would prefer to use tiered storage behind their VMs to facilitate the storage of application and protection data on "separate storage tiers." (Since "tiered storage" is a marketing term that means different things to different people, I am not sure if this finding passes muster with my scientific survey test.)
Finally, and I am not sure whether this would be different in a non-virtualized environment, the top two backup challenges cited by responders were "too much I/O overhead" (41 percent) and "scheduling backups" (46 percent). These percentages seem to align with surveys about tape backup generally.
It is good to recall that, since the preponderance of respondents said they were deploying VMware, these pain points correlate most significantly to perceived deficits in the VMware solution. One that Derrington saw fit to emphasize was VMware's "closed" SAN pathing scheme, which "creates more challenges than Microsoft's multipath I/O." He added that half of the respondents want to connect their Fibre Channel fabric-attached storage to their VMware hosting platforms and proffered Veritas Cluster Server as a means to facilitate what VMware does not.
There you have it: a full accounting of Symantec's 100-person survey. If conditions are as the Symantec research suggests -- that storage is harder to manage and data more difficult to protect in an X86 virtual machine setting -- then this should be triggering alarms and flashing lights everywhere.
This observation aligns with the press release traffic that flowed from the show. I'm not sure of the exact number, but there was certainly no shortage of vendor press releases that underscored the deficits of server virtualization from the management and protection point of view, albeit more palatably by emphasizing the "value add" that storage hardware vendor ABC or software vendor XYZ was contributing to the server virtualization story. In the final analysis, this might be the most important insight to be derived from all of the information exchanged at that conference.
Your views are welcome: jtoigo@toigopartners.com.