BACKGROUND IMAGE: iSTOCK/GETTY IMAGES
There are some issues with the way VMware and Microsoft flash memory works. Flash is an up-and-coming technology, and many people believe they can deploy it in their server in a caching role ahead of the disk they may use for primary storage. But flash is good for reads, and not so great for writes. You have to be very conscious of flash wear, and take advantage of its strengths. What virtual storage vendors usually do is write coalescence.
Flash vendors use DRAM, which isn't wear-sensitive. They aggregate all the writes into the DRAM cache to make a smaller number of large writes into their flash cache. That preserves the integrity of the flash and protects it against flash wear. VMware doesn't use any pre-cache in write coalescence. Instead, it sends many small writes from all the workloads you've put on the stack -- but this hammers the flash card and could cause it to burn out.
Hyper-V poses a slightly different flash wear problem. It uses deduplication on its storage, but writes original files. The original data is written to the flash cache as a bunch of small writes. These writes hit the cache a second time when Hyper-V runs the deduplication algorithm against it. So Hyper-V flash will burn out even faster than VMware will. So neither vendor uses flash technology very effectively. In addition, flash cards aren't inconsequential in terms of their cost.
Starwind Software, by contrast, uses flash very well. And DataCore does write coalescence. All the other third-party, hardware- and hypervisor-agnostic vendors also use flash very well. That's a good differentiator between their products and the products of the main players.
A look at the true cost of flash storage
How IBM is working to address flash memory endurance
Four ways to extend the life of your flash storage
Have a question for an expert?
Please add a title for your question
Get answers from a TechTarget expert on whatever's puzzling you.