Page 1 of 3
Early in this century, the emergence of the desktop virtualization was hailed as the second coming of centralized computing. As recently as last September, Gartner reported that more than 80 percent of enterprises had a virtualization program or project in place, but that only 25 percent of server workloads would be running in virtual machines by the end of that year.
What's holding things up? With the promise of instantaneous deployment, centralized management, user mobility and simplified data protection, why haven't more companies leaped into the VDI void? One problem has quite literally has been the scale-up speedbump.
Sure, your customer's servers cranked along nicely through the pilot phase and straight through the implementation of about 200 users. Then the roof fell in and complaints rained down about slow performance, sporadic availability, frequent crashes and data loss.
The cause is often I/O bottlenecks, which can quickly overwhelm a VM server when large VDI systems are deployed on a one-image-to-one-user basis. "Desktops represent the next explosion of I/O; they're more random and less predictable," said David Bieneman, CEO and founder of Liquidware Labs, which develops platform-level VDI solutions. "Desktops are used to having their own disks, not sharing in a pool."
So naturally, there will be wait times as large numbers of virtual systems queue up to read and write data, Bieneman said. "And desktop users are less forgiving of wait times," he said.
Liquidware Labs products examine user and application traffic patterns over a period of time and identify the most I/O-intensive applications. These can be isolated and deployed on dedicated storage and network channels designed to handle high I/O traffic levels while VM servers and file systems dish out a small number of desktop master images.
"Profiling data center and desktop environments with our approach allows you to design around observed storage I/O patterns," said Bieneman. For VDI systems, this information allows systems designers to more precisely match shared storage capacity and type to what is needed to maximize performance and minimize cost. "Administrators don't have to store one-to-one per user; they can store perhaps 64 users to one storage space."