Long ago, when servers still came one to a box, "sysadmins" spent all their time running from one machine to another, with boxes of tools and utilities designed to squeeze out every bit of performance and stability from physical servers. Now, virtual servers outnumber the physical in most data centres. And neither budgets nor toolboxes are over-provisioned with resources for fine tuning virtual infrastructures.
Companies expanding beyond the pilot phase and moving into large scale server or desktop virtualisation need to realise that utilities from third parties, not just the platform vendors, are what will help them make virtual infrastructures as stable as the real ones, some analysts say.
"One of the most common misconceptions among the non-technologist crowd is that once you buy your virtualisation platform, you are done with your software purchases," warns Greg Shields, who writes extensively in books and blogs on the details of virtual infrastructure implementations and serves as partner and principal technologist at Concentratedtech.com.
"When you virtualise an environment you add in dependencies no single human can do a truly good job of understanding," Shields says. "On a physical server a network problem is probably related to the card. On a virtual server, the card is virtual, so the problem could be it's not getting enough processor power, or the storage performance. You need very broad-based tools that will address those metrics."
The fight between HP and Dell to acquire virtual storage optimisation vendor 3Par shows how important management and optimization products are in keeping virtualized infrastructures running, according to Chris Wolf, senior analyst at Burton Group.
"This year especially we saw a lot of big organisations virtualising serious enterprise applications," Wolf says. "When you have mission critical apps virtualised or in the cloud, diagnosing application problems and optimising performance in the virtual environment becomes very important."
It's impossible to say which utilities or ISVs offer the best tool for every environment, Shields says. But five specific types are particularly important to getting virtual infrastructures humming right now.
1. Capacity management
"Virtualisation is taking what became sort of an also-ran activity, capacity management, and showing why it's really a critical step," Shields says.
Multi-processor, multi-core servers and acres of RAM made planning for server capacity almost moot, Shields says. With virtual servers, however, the question isn't the power of the server, it's how that capacity is doled out to specific workloads on specific virtual machines, and monitoring the performance of the VMs to make sure all the resource demands are satisfied.
"It goes beyond not being able to automate anything until you know what you have," Wolf. "Without capacity management you don't know what a particular service is costing the organisation and that makes it harder to build out your infrastructure as a service."
VMware's vCenter CapacityIQ is effective at identifying utilisation gaps, Shields says, but there are plenty of other options. These range from old school IT favourites retooled to cover virtual as well as physical, such as BMC's Capacity Management and HP's Insight Dynamics, to purpose-built virtualisation management tools from VKernel, VMTurbo and Embotics.
2. Performance optimisation
Performance problems in physical servers are relatively easy to spot because most functions are associated with a specific component. Swap it out and you're good to go.
"On a virtual server a performance issue could be related to spindle contention in storage, an oversubscription of RAM, and undersubscription of RAM, under or over subscription of processors, bandwidth utilisation; a whole series of dependencies that make it hard to put your finger on the problem without a deep analysis of what's going on inside," Shields says.
Hyper9, for example, offers a set of tools called Hyper9 VEO (Virtual Environment Optimisation) designed to discover all the VMs in an infrastructure, all the applications running on them, the relationships between the applications, VMs and physical servers and to collect data on performance, configuration and capacity.
Those capabilities were de rigeur in the physical world, but are still uncommon in the virtual, Wolf says. ISVs such as Akorri, Netuitive and VMware's CapacityIQ are also making good progress on performance optimisation tools, he says.
3. Storage management
Storage continues to be one of the most consistent source of headaches for virtual infrastructure managers, Wolf says. Converting physical servers to virtual requires more back-end storage space, a problem exacerbated by VM sprawl, and not even virtual storage systems are typically designed in ways that make it easy for virtual servers to run at their best, Wolf says.
"Companies have been able to plan their CPU and memory density, anticipate boot storms that generate a lot of I/O, but they haven't always been able to optimise tiered storage for virtual servers, or do things like queue data locally so you aren't pushing as much data through the pipe," Wolf says.
Three-year-old ISV Virsto tries to address VM storage problems by reducing the amount of disk space used for VMs by eliminating the need to store the same data for 100 golden images, and improve performance by reducing the number of data packet collisions VMs generate at storage I/O busses by not coordinating their timing the way a single server would.
Akorri's BalancePoint virtual appliance also does a good job of mapping virtual servers to physical storage and, keeping real time data on performance and capacity and notifying IT if a problem develops, Wolf says. Netuitive has also done a good job with its self-learning technology to detect and flag failures, he adds. VMware storage products include storage I/O management, performance and resource managers and backup, among others, but the company spends as much effort remaining neutral in the storage market so as not to seem too tied to parent company EMC. EMC, on the other hand offers an array of hardware and software products designed to optimise storage for virtual environments; to simplify things, it announced in August an integrated product set called Unified Storage.
4. Virtual enterprise management suites
Not surprisingly, a lot of the physical-IT-management vendors have been eager to expand their reach into the virtual world as well, and have done so very effectively, Wolf says. VMware partners BMC, CA, HP and IBM have all been making creditable forays into virtual enterprise management, even against VMware's claim that its vCloud Director is purpose-built for the environment, Wolf says.
"You have a lot of very sophisticated capabilities coming in from these platforms, like the service automation suite HP got from OpsWare," Wolf says. "There's more pull there from that level of capability and because VMware knows it's not going to be a device management company, so it's not reaching down where some of the enterprise ISVs already go."
5. Desktop virtualisation planning and management
"Virtual server environments are an order of magnitude more complex than physical server environments because of the additional ecosystem they add to the physical one," Shields says. "Desktop virtualisation adds even one more ecosystem and a lot more."
Virtual desktops can also be delivered in more ways than virtual servers, ranging from full-on VDI in which each user gets a dedicated VM with a single OS running on a backend server, to virtual applications that can be viewed from almost any machine, Shields says.