dcsimg
Read more on "Server Virtualization Spotlight" »

Solid State Storage's Surging Impact on Server Virtualization

By Paul Rubens (Send Email)
Posted August 8, 2012


When you think of server virtualization you probably think of big grunty physical hosts and powerful but lightweight hypervisor software. But what's not quite so obvious is the key role that NAND flash technology plays in server virtualization.

It's becoming increasingly apparent that, like mustard on a hotdog, a little dash of flash can make virtualization much, much better. Or, as Mark Peters, an analyst at Enterprise Strategy Group, more soberly puts it, "A judicious use of a small amount of solid state storage can improve performance."

Given that the performance of solid state storage devices far exceeds that of spinning disks you'd think that it would make sense to go for more than just a judicious use of solid state. To go all mustard and no hot dog, so to speak.

Virtually Speaking But there's really very little point in that, Peters believes, "unless and until all storage is the same price, or free."

Essentially, virtualized applications that require only solid state storage at this point are corner solutions, to use economic-speak. These applications do exist, but it's more likely that the optimal solution, in terms of price and performance and so on, will be a balance of the two.

Amazon clearly believes that databases in public cloud computing require corner solutions, and its EC2 service is now offering a High I/O Quadruple Extra Large (hi1.4xlarge) virtual machine instance type. While this might at first sound like some kind of candy bar from Charlie and the Chocolate Factory, a hi1.4xlarge is in fact an EC2 instance equipped with 2 1TB SSDs.

For 4k random reads the hi1.4xlarge offers 120,000 IOPS when using a paravirtualized VM, or 90,000 with a standard one. "With 15K RPM magnetic disks you will see a bit over a hundred IOPS at best," says Werner Vogels, Amazon's CTO. "Even though SSDs are still more expensive from a storage point of view, they are a much more cost-effective solution from an IOPS point of view."

Corner solutions aside, with server virtualization so common it means there are many, many virtualization hosts that would perform better by being connected to some form of hybrid storage rather than plain old fibre channel and SAS arrays. That explains some of the hybrid arrays that are now available, and it also helps shed light on the surge in popularity of specialized hybrid storage appliances and other devices that are optimized for virtualized environments.

Hybrid storage appliances from the likes of Tintri and Nutanix bring solid state storage close to the physical hosts, where they can store and supply from solid state storage the hot data when it's needed at high speeds, while leaving the rest on cheaper, lower performance hard disk drives.

It also explains what vast caches of solid state storage are doing inside physical server virtualization hosts. Companies like OCZ, Fusion-io, EMC and LSI are promoting the judicious use of solid state memory attached directly to the system bus via the PCIe interface as well as using their proprietary software to make the flash storage available to all the VMs running on the host.

OCZ's software is called VXL, Fusion-io offers io-Turbine, EMC has VFCache and LSI markets its Nytro XD solution. The idea behind all of it is the same, though: to speed up application performance and relieve I/O bottlenecks by letting VMs talk to the flash cache, and letting the flash cache write the data to external storage systems in a more optimized, sequential manner.

The idea of storage tiering itself is not new, but what is surprising is just how far and how fast tiering using flash has become commonplace in environments where server virtualization technology is used. One reason is that prices have fallen rapidly over the last few years, and during the same period storage performance issues have become more pressing.

Flash technology prices are likely to continue to fall for some time to come, and as caching algorithms become even more efficient, it looks like we'll increasingly want to splash some cash on flash.

Paul Rubens is a technology journalist and contributor to ServerWatch, EnterpriseNetworkingPlanet and EnterpriseMobileToday. He has also covered technology for international newspapers and magazines including The Economist and The Financial Times since 1991.

Follow ServerWatch on Twitter and on Facebook

Page 1 of 1

Read more on "Server Virtualization Spotlight" »

Comment and Contribute

Your name/nickname

Your email

(Maximum characters: 1200). You have characters left.