Balancing the Virtual Load
As computer hardware becomes increasingly more powerful, organizations are reaping huge cost and energy savings from replacing physical machines with virtual ones.
But whether you're replacing several physical servers with virtual machines running on a single host, streaming virtualized desktops to thin clients, or delivering virtual applications, if you don't balance the load on your resources, your entire network can come screeching to a halt.
Think of load balancers as traffic cops, says Lew Smith, director of business development and virtualization solutions for consultant Interphase Systems. The devices keep the bits flowing smoothly to optimize performance, manage resources and ensure redundancy.
“Without load balancing, you'll get networking bottlenecks and reduced performance from apps,” says Smith. “It's like comparing a five-lane highway to a two-lane road. Both can handle the same number of cars, but the two-lane road will take a lot longer.”
Like the machines themselves, load balancing can be physical or virtual. Smith says most enterprises deploying virtualized environments use a combination of both: dedicated hardware appliances to manage data traffic coming in from the Internet or across internal networks, plus load balancing software to route data efficiently between physical layers and virtual ones.
The Andover Public School district in Kansas moved to a virtual server network four years ago, says Rob Dickson, the district's director of technology. The main reason? Andover's data center couldn't keep up with the district's growth; they needed to virtualize simply to generate enough computing power.
“If any of our servers get beyond 90 percent utilization, SolarWinds notifies me, and we Vmotion those virtual machines over to a new physical server,” says Dickson.
Disaster recovery provided another compelling reason for Andover to go virtual. When you live in the heart of tornado country, you need to be prepared. The district set up a DR site using Data Robotics DroboPro storage appliances, which take a snapshot of each virtual server every 15 minutes. If a twister strikes, Dickson and crew can simply pick up more or less where they left off, without having to restore from a backup.
650%: Anticipated growth in enterprise data over next five years
Campbell High School District in San Jose, Calif., uses Citrix XenDesktop to deliver virtual desktops to its students. In this scenario, all the heavy computing is done on the back end by powerful HP BladeSystem servers, which continuously stream images to each student's machine. Even heavyweight apps like Adobe CS4 or streaming video run smoothly because Citrix is sending only about 300 kilobytes of data to each machine at a time, says Charles Kanavel, director of technology.
If you had had 30 students watching YouTube in a nonvirtual desktop environment, they'd all slow down at once because the network switches would get overloaded with data, he says.
Now load balancing happens automatically on the back end. The Citrix software creates a pool of virtual machines on hot standby; as each student logs on, they're routed to the blade with the most resources available. If processing demands exceed what's available, XenDesktop spins up new VMs.
Another plus to virtual desktops is that it closes the technology gap between schools with extensive funding and those without, says Kanavel. “In our district we've got schools on both ends of the spectrum,” he says. “When you're using virtual desktops, you get the same experience no matter what your socioeconomic status or what equipment you're using. It really bridges the digital divide.”
Lifting the Load
Here are three tips for managing load balancers in a virtual environment.
• Think big picture. Load balancing needs to happen everywhere bits flow – across internal networks, between applications, between physical and virtual layers and from storage to memory and back. Always provide at minimum one level of high availability and redundancy at every layer.
• Measure twice, architect once. Monitor how data flows across your physical environment before you implement a virtual one so you can allocate enough resources for peak traffic needs. Without sound data, captured over a period of time, it's nearly impossible to architect a solution that can handle peaks and valleys of performance intelligently.
• Choose the right level of automation. If you have sudden spikes in demand – such as “boot storms” when several students log on at the same time – set your virtual infrastructure to intelligently seek out resources and scale as needed. Otherwise, opt for a more balanced setting or even manual override.