Hypervisor Hype

For all of the hype around hypervisors - the software technology touted to optimize server infrastructure by virtualizing host machines, you'd think that we would see some signs of change. It's about time for a reality check.

At the beginning of 2010, industry analysts pegged market penetration of hypervisors well below what the leading vendors had expected: only about 15 to 20% of all servers had gone virtual.

More to the point, the preponderance of workload being moved from physical to virtual servers remains mainly of a class best described as "non-mission critical" - that is, file servers and low traffic web servers. And technically speaking, neither of these two types of servers require a hypervisor at all.

• File servers, for example, can be consolidated using that old reliable virtualization engine, the file system, that has existed for years. If an old file server gets a bit long in the tooth, administrators can just package up its files into a new folder on a bigger box. No muss, no fuss.
• Low traffic web servers can also be effectively hosted and shared on a single box, especially when a common front end manager (like Plesk) is used.

Where hypervisor technology has been used to consolidate these workloads, other problems have cropped up - especially with respect to storage I/O performance. We would be interested in learning the total revenue value of the ecosystem of vendors whose products have been developed primarily to follow behind hypervisor deployments and to fix what they break. These ancilary costs are helping to break the back of the revolution that server virtualization was supposed to usher in.

The Deafening Noise, The Deafining Silence

The trade press has, until recently, done a pretty poor job of covering the downside of server virtualization, contenting themselves instead to rake in the desperately-needed advertising revenues of the key players. Said one editor of a prominent trade in late 2008, "Are you kidding? The virtualization guys are buying all of the ads. We won't write anything negative about them." So much for the separation of church and state at that publication!

But the real stopper on the free flow of useful information about the vicissitudes of virtualization has generally been the users themselves. Some fall into the class of "true believers" who think that the wrinkles in server virtualization wares will be ironed out over time. Another group is simply too intimidated by the loud noise around virtualization to raise any doubts or concerns - either fearing that they will be characterized as reactionaries or flat worlders who fear inevitable change, or worse that their IT acumen will be questioned when they report the shortcomings of their experiments.

Business managers want consolidation as a part of cost-containment. They want fewer servers, lower administrative and management staff levels, lower electricity bills, and greater resource utilization efficiency from the gear they are buying. Some have shot themselves in the proverbial foot by buying into the idea that the same hardware that is being consolidated should be used to host virtualized applications - even if its CPU brain predates the generation of processors updated with multi-tenancy code.

Meanwhile, server sprawl has become virtual server sprawl. Storage area networks (not really networks, but that's another story) are slowing to a crawl. Networking expense is actually accelerating, rather than decelerating. And free hypervisor licenses delivered with new gear must be upgraded at high cost to get the sexy functionality described in the vendor brochure.

Let's not forget that just developing and implementing a server virtualization strategy is typically an expensive endeavor. Software tools have been created to help identify what applications might perform adequately in a virtual server setting - and which ones to avoid virtualizing. Consulting services are available to prevent "seat of the pants" deployments from becoming nightmares. Network and storage analysis services are available to help identify the collateral impact of server virtualization and to identify what parts of infrastructure will need to be redefined - or scrapped in their entirety - for the strategy to have a snowball's chance of yielding any real business value. All of these products and services represent a potentially huge cost, over and above the hypervisor software license and a rig to run it in.

Disaster recovery and security requirements must also be considered when going down the virtual path. Some of the data protection processes you operate today will need to be "refreshed" because they don't work at all well in the new virtual reality. In a few cases, vendors are trying out the line that disaster recovery is no longer needed at all: hypervisors fulfill the need with integral high availability capabilities. Missing are the statistics showing the rate of failover failures between servers in a common subnetwork. Concealed from view is the fact that some of the HA capabilities aren't integral to the hypervisor product at all, but are provided by third party software companies under the covers.

Options, Anyone?

Missing or entirely dumbed down in the current hypervisor hype is the most basic and fundamental of analyses: options for server consolidation - whether they involve virtualization or not.

• A Linux OS server, for example, makes pretty efficient use of its resources and its OS supports multiple concurrent accesses pretty well. So does the new Windows 7 64-bit OS. Does a hypervisor increase this efficiency so significantly that it is worth the hassle? 
• Consolidation via virtualization requires that every server that might potentially host a resource demanding application be configured with the number and type of network and storage I/O connections to facilitate the operation of that resource hogging app should it come to operate there. Doesn't this drive up the cost and complexity of server and other infrastructure configurations? Is it more or less costly than the alternative of non-virtual servers?

• Does virtualization increase risk? It seems to, given that the lack of insularity between guest machines usually results in all guest applications failing if any one application fails. While this doesn't happen in a mainframe logical partition (they've had 30 years to work out the bugs), it would seem to be a risk multiplier in virtual server environments.
• When we were told that that old "patch and fail" scenario that blessed us over the years with billions of "Dreaded Blue Screens of Death" in the Microsoft server world was a thing of the past with virtual servers, did someone forget to mention that VMware has a Dreaded Pink Screen of Death that many shops encountered in 2008 when an ESX patch caused tribulations for early adopters of the hypervisor around the globe.

• Last (but most certainly not least), we wonder when someone will work out a real cost model for server virtualization that includes all of the expenses that offset the purported advantages, including pre-virtualization activities, software licenses and renewals, equipment expenses including networking and storage costs, administrator training costs, third party monitoring and management software required to make the platform run as advertised, and anything else from real world experience that can be thrown into the mix. We would not be surprised to see actual cost significantly greater than the numbers that predominate hypervisor hype.

To be sure, server virtualization has its place, even with all of its current warts. Our test labs use it often to build up and quickly tear down servers we are using in some of our work. But we don't host mission critical workload there.