alphaspirit - Fotolia
My good friend and respected colleague, Marc Staimer, president and founder of Dragon Slayer Consulting in Beaverton, Ore., did a podcast on SearchVirtualStorage that got my attention. It wasn't so much what Marc said that caught my attention -- which was, as always, sage advice from a Jedi Master -- but the headline wrapped around the podcast and transcript: "VMware hypervisor swap-out means loss of data storage features." Whether it was the headline writer's intention or not, there's a lot of nuance in that title that I want to unpack in this column.
Virtualization perception vs. reality
Starting at the beginning, the title suggested that folks might be contemplating a change in hypervisor vendors, a change away from VMware. Late last year, backup software vendor Veeam Software commissioned a survey that queried 578 firms in the U.S. and Europe, and found that something like 67% of U.S. firms, and up to 79% of firms in France, Germany and the U.K., used VMware. The data seemed to suggest that in companies with more than 1,000 employees (those sampled in the survey), VMware ruled the roost.
Interestingly, the same survey showed that perceptions of VMware server consolidation success were greater than reality. Companies believed they were putting approximately 10 virtual servers on every physical server, though the reality was about half that number on average.
The really remarkable stat from the survey was the 59% of respondents who said they would likely be changing their preferred hypervisor vendor this year. Most were unhappy with the licensing fees charged by VMware, while some cited the improved maturity of competing products.
Digging deeper into the minds of server virtualization planners, I looked at DataCore Software's Third Annual State of Virtualization Survey last year and discovered that storage-related costs were cited by 339 of the 477 respondents as an obstacle to realizing value from server virtualization initiatives. And 308 said storage inefficiency was mitigating their efforts to achieve the performance levels promised by hypervisor peddlers.
So, the first part of the SearchVirtualStorage headline seemed to have merit: There is, in fact, an incentive to change hypervisor vendors. And since VMware has the lion's share of the market, we can safely assume that a lot of this hypervisor-swapping action will involve changing VMware stuff for other brands.
VMware storage features overhyped
The second part of the headline, however, is what bothered me the most. There was a not-so-subtle suggestion that if you changed out VMware you would lose some important storage capabilities -- or, as the headline said, changing hypervisors would mean "a loss of data storage features." Looking back, it seems to me that VMware storage-related "features" have largely been bugs introduced by, or break-fixes necessitated by, very poor architecture.
VMware did a terrible job from the outset with storage I/O. The legendary I/O logjams created by VMware's hypervisor had nothing to do with the (in)efficiency of the storage beneath the virtual server. Looking at the situation with a performance meter, you could see the processor running hot and I/O queue depths of zero. The hot-running CPU meant the slowdown in I/O processing was linked to bad code in an ESX microkernel, while the non-existent queue depth meant the storage infrastructure was having no difficulties handling the trickle of I/O that was managing to find its way into the host bus adapter or network interface card on its way to the storage.
Workarounds were introduced, including an effort to offload "up to 80 percent" of the storage I/O to the intelligent array controller, to make the hypervisor seem more storage-aware. Activities like mirroring could be better accomplished on a proprietary array controller said the VMware folks when they introduced the vStorage APIs for Array Integration (VAAI). So, in effect, they were exposing their customers to a hardware vendor lock-in while at the same time introducing unapproved commands into ANSI T10's holiest of holy standards, SCSI. This was repeated a few months later with a second-generation VAAI, introduced once again without standards approval or notification of the hardware vendor community.
VMware succumbs to SCSI standards
For the record, the company has since submitted some of its "innovative SCSI commands" to ANSI for review and approval, claiming they're "driving the standards process." However, I've been told by knowledgeable sources that this newfound spirit of cooperation was driven less by an affinity for open standards and more to appease large customers for whom standards-compliance is a must.
Since then, we've seen the introduction of de-evolutionary architectures such as VSAN and server-side storage (aka direct-attached storage), and earnest fanboys of the hypervisor are waiting on the company's first hardware-centric OEM play (a productization of VSAN with various hardware partners) that had been code-named Marvin. As currently described, Marvin will provide a VMware-branded server-side storage kit that, like the company's other software-based VSAN technology, works only with its hypervisor stack. Given that larger firms will likely have a multiplicity of server hypervisors as well as some workload running on non-virtualized servers, we're returning to isolated islands of data -- a problem we tried to fix with SANs starting back in the late 1990s -- and infinite levels of data replication and mirroring to support failover and vMotion in the absence of a persistent storage mount from a shared, well-groomed and well-managed storage infrastructure.
I'll repeat: Isn't VMware storage well-managed? vSphere uses REST, doesn't it? Yes and no. There's REST in there, I was told recently by a savant of RESTful management programming, but it's hidden behind several proprietary API layers so as to be barely useful.
Bottom line: From where I'm standing, you aren't giving up very much at all, at least in the way of storage features, if you abandon VMware for an alternative server hypervisor or for no server hypervisor at all.
About the author:
Jon William Toigo is a 30-year IT veteran, CEO and managing principal of Toigo Partners International, and chairman of the Data Management Institute.