NEW YORK -- Virtualization and its effect on storage was a big topic at Storage Decisions last week, as attendees at the show received tips for managing storage with virtual desktops and servers.
Brian Madden, virtual desktop analyst and founder of BrianMadden.com and ConsumerizeIT.com, offered advice on running
In his keynote address on the second day of the show, Madden said cost reduction is the wrong reason for deploying VDI. He said VDI will not deliver the consolidation and cost savings benefits that organizations gain from server virtualization.
"VDI is not about saving money," he said. "It's about enabling great new features."
Madden said the benefits of a VDI environment are that it allows users to access their desktops from any device, anywhere, and results in higher availability. Downsides include the fact that it doesn't work offline and performance and access can suffer in places with limited bandwidth.
Madden also explained that implementing a VDI environment for security reasons may not be the best use of an organization's resources. Commonly listed VDI security benefits include locking down desktops, eliminating administrative rights for end users, and eliminating end users' ability to install their own software. But Madden said IT administrators have had these tools for years, and without all of the time and money costs of implementing VDI.
To get the feature benefits of a VDI environment, Madden said IT administrators should focus on storage system performance to ensure the end-user experience doesn't suffer because of the shift to a virtual environment.
He pointed out that end users will notice when VDI storage is inadequate. "In reality, users want their IOPS," Madden commented.
Madden said a typical desktop or laptop SATA hard drive provides an average of 50 to 70 IOPS, and admins should at least match that performance within the VDI environment. He reminded the audience that a virtual desktop environment typically employs an equal amount of read and write operations, which is far more write operations than a typical virtualized server.
And if you are throwing dozens or hundreds of VM-based virtual desktops onto each physical host, you have to be prepared for the "I/O blender" issue created by having all the read and write operations streaming out of multiple virtual desktops at the same time.
"You start adding a lot of users to servers with VDI and it's not going to end well," Madden said.
Marks took a deep dive into virtual server storage with his session, How vSphere 5 Changes Storage Configuration, Maintenance and Management.
In vSphere 4, Marks said DRS relied on CPU and memory usage to balance workloads using vMotion, ignoring storage resources in the calculations and virtual machine (VM) movement. Version 5 adds available disk space, I/O load and latency to the mix for initial placement, and load balancing. Version 5 also allows the creation of anti-affinity policy rules to make sure VMDKs either stay in the same data stores or are separated. Using SDRS, users can also aggregate multiple data stores to create clusters.
"The downside to virtualization is sprawl," Marks said when describing the benefits of creating data store clusters. Marks illustrated environments that he was called in to consult on that went from 50 physical servers to 400 VMs. With vSphere 5, those virtual disk data stores can be clustered and managed like a pool of storage for increased performance and reliability.
Marks said he found out the hard way how valuable vMotion is because his consulting fees took a hit due to the "vastly simplified" data migration in virtual environments it brought.
"My least favorite feature in VMware is vMotion," he joked.
VMware made some changes to accommodate increased usage of Storage vMotion. Mirror Mode replaced Changed Block Tracking (CBT) and allowed writes to in-progress Storage vMotion moves, and it supports migration of vSphere snapshots and linked clones.
Marks said other valuable virtual server storage features include Atomic Test and Set, which allows block-level file lock so you can pile more VMs on each physical host; faster cloning with Clone Blocks; Block Zeroing; and Out of Space Thin Provision Stun, which suspends VMs when a thin-provisioned disk runs out of actual capacity instead of crashing the physical host.