The data center manager often works under the onus of executive management's wishes. The CEO demands responsiveness, flexibility or whatever buzzword the vendors are promoting. The chief financial officer wants to do it for less. And the chief technology officer needs more uptime, all the nines you can get.
Commodity server consolidation through virtualization is becoming a viable tool to reach those goals.
Virtualization was developed on the mainframe decades ago. The reason for the technology's resurgence is the development of virtualization for industry standard servers (x86 and AMD). According to Raghu Raghuram, senior director of strategy and market development for Palo Alto, Calif.-based VMware, this isn't news. "But many people still need to be educated," Raghuram said.
Virtual machines (VMs) are guest operating systems and applications, running on a virtual layer on top of the physical hardware. According to Raghuram, virtualization is a broader concept than partitioning. VMs are autonomous from the notion of physical resources. A VM considers itself separate from the physical entity, whereas a partitioned server is sharing resources across one piece of hardware.
"VMs are encapsulated as a single file, one that you can copy and paste to any other physical hardware," said Suman Kumar Singh, a Dell systems engineer, in a presentation at LinuxWorld Conference & Expo last week. "Because it is a single file, it is isolated from other VMs."
During the data center sessions at LinuxWorld in Boston last week, vendors promoted the use of VMs in cluster architecture. According to presenters, clustering two or more systems to work together to complete a single task -- such as load balancing, high performance computing or high availability -- is an area where data centers can use VMs to gain maximum efficiency from physical hardware.
By clustering VMs, instead of clustering physical servers, data centers could reach new levels of consolidation and availability, according to Singh. VM clusters can be applied within one physical machine or clustered within separate physical servers. According to Singh, a hybrid, physical-to-virtual structure protects against hardware and software failures, while still reducing physical redundancy.
Other structures include creating a virtual pool of servers by dedicating one machine running only backup VMs. In this instance, when a physical server fails, applications transfer to a VM from the pool. Because the VM is a single file, it can be cloned and redeployed to a new or repaired physical server in minutes.
"There are a lot of challenges in this area; it's not very mature," Singh said. "One way to make things easier is to run a homogeneous environment. We're offering a complete solution, running VMs, OS and hardware on a single source."
In other words, every standard server vendor is coming out with its own virtualization model, and is doing it in different ways. Therefore, there is no standard, and data center managers will have to use a single vendor for this type of consolidation.
Vendors banged on the virtualization drum at LinuxWorld. But what is the motivation to promote it now? According to Tony Iams, analyst with Ideas International, a Port Chester, N.Y.-based research group, it has to do with timing. With the lifecycle of the glut Y2K servers winding down, there is significant demand for hardware, and consolidating resources to VMs is becoming a mainstream option.
"Consolidation doesn't happen overnight. You wait for a refresh," Iams said. "You try to get the most utilization. Instead of buying 10 servers for 10 applications, you buy fewer servers and use virtualization."
And while consolidation and high availability are major concerns for data center managers, the real story, according to Joe Clabby, vice president and practice director with Boston-based Summit Strategies, is provisioning -- automatically distributing workflow so applications get the resources they need across VMs. According to Clabby, IBM Tivoli, Hewlett-Packard and VMware offer programs that automatically provision VMs.
"Reallocating hardware to deal with unexpected traffic can take a lot of time. But with virtualization, customers can create images of VMs and deploy them on other physical machines in minutes. You can even use a server template to make it more standardized. There are also programs available to automatically balance loads," Raghuram said.
"Consolidation and high availability are benefits of virtualization, but payback will come from provisioning. These are small benefits compared to the costs you can save from not having to allocate people to manage your computing resources," Clabby said.
Let us know what you think about the story; e-mail: Matt Stansberry, News Editor
Dig Deeper on Data center capacity planning