BACKGROUND IMAGE: iSTOCK/GETTY IMAGES
Desktop virtualization and VDI bring security and efficiency benefits to the table, but the inconsistent nature of Windows client I/O is a big problem in many shops. Luckily, all-flash storage arrays can help.
Booting virtual machines (VMs), updating security, scanning for viruses and other common desktop tasks can create disk access storms. The most common problem for VDI is that reads -- in terms of these boot storms, reboot cycles and other times when users launch applications -- are extreme on storage IOPS for short periods of time. Then, the virtual desktops typically switch to a steadier read/write scenario throughout the day where the effect on I/O is lessened considerably.
If you're looking for ways to reduce those storms, you might try tactics such as doing security scans outside the VM, standardizing desktops on a single image, thin provisioning, limiting applications and more. But while these solutions are all well and good, they still don't save you from certain periods of I/O slow down. When that happens, users panic and they do things that make the overall problem worse, such as rebooting en masse. And when users aren't happy with their desktop experience, it can slow down a VDI deployment or even stop it in its tracks.
Get flash, ditch (some) I/O woes
The most effective way to resolve I/O issues is to turn to all-flash storage arrays. Although flash can't get rid of boot storms or related surges, it can handle those problems better than traditional options.
Solid state drives, on-board flash and hybrid flash with spinning disks are several different options, but you will need at least some flash for VDI to work properly.
When it comes to VMs and the OS and application files they depend on, I think an all-flash array is the best because many partial-caching systems yield unpredictable results. Transitional tiered-flash implementations can be difficult to work with, especially if they sit on existing storage arrays.
There are many more companies and approaches to flash storage than there were even two years ago. The big enterprise storage vendors aren't the only ones in the game; there are plenty of start-ups aiming to provide their own take on all-flash storage arrays based on commodity parts and a better management layer. Some plug into your current infrastructure, but many of the more promising options are singular boxes that don't bell-and-whistle you into a whole storage system. Instead, they provide flash and flash only.
You'll want to keep a critical question in mind as you move to flash: How does the storage deal with spiking reads and steady writes? That huge disparity is why I am working on moving an entire virtual infrastructure to an all-flash array.
The current hybrid caching and tiering system in my shop isn't getting the job done for me or my users. It either runs too slowly when it comes to critical times such as boot storms, or it sends storage to spinning disks in an inopportune way -- with a batch-job mentality and caching warm-up period that doesn't react quickly enough.
All-flash storage arrays' cost and value
One area that is still an obvious pain point is the cost per gigabyte. Flash is still much more expensive, but in the grand scheme of enterprise storage and 15,000 spinning hard disks, it's not necessarily that much more costly than what you pay today.
Buyers should also be aware that when a vendor advertises a 40 terabyte storage capacity for example, that number is based on specific deduplication and compression algorithms that the company does not usually disclose. In addition, vendors use different types of flash storage as the base technology, such as SLC and MLC. But using those components as a measure of performance is no longer accurate because vendors enhance that base hardware technology with software. All flash-storage products are built out of the same components, but the software that vendors put on that hardware is what really makes one product different from or better than another.
The question will then become "How fast does the array really need to be?" When it comes down to brass tacks, you need IOPS. If you can't give users the IOPS they get on the laptops and desktops they have today, then the trickery of other options only slows the bleeding rather than closing the wound.
You can keep Word files and other user data on spinning disks on another NAS or SAN, but that core OS needs to be on flash and it needs to be fast. They key is having enough IOPS for each individual user to boot up as if they were on a dedicated machine. If you can't provide that, then don't even consider the hardware.
All-flash arrays offer enterprise functionality
Hybrid storage vs. all-flash arrays
Are all-flash arrays best?