This week was a heyday for VMware with VMworld taking place in San Francisco, a slew of new product announcements, and a healthy stock bump on Monday. It is almost impossible to avoid the Virtualmania! taking place...so we won't try to.
Dave Hitz at NetApp wrote an interesting post today titled Why Run VMware over NAS? He has a few great points about managing files versus LUNs, and even finds a way to work in some praise for Chuck Hollis at EMC. Nice to see everyone agreeing on the same thing...and getting along for a moment.
It is encouraging to see two thought leaders promoting the simplicity and functionality of NAS for virtual environments. Most people tend to agree that NAS is easier and more cost effective than SANs for modern data center architectures. But NAS still has a lingering performance stigma that often remains high on administrator minds...to the point where many people feel they need to consider alternative solutions such as Fibre Channel SANs.
The stereotypes about NAS and SAN are longstanding...to the point where two of the world's larger IT companies, Oracle and NetApp, still have to showcase how their products deliver on performance. For a humorous, but technically strong debate on the subject of NFS performance take a peak at Manly Men Only Deploy Oracle with Fibre Channel - Part 1. Oracle Over NFS is Weird.
Now back to the VMware discussion. As Dave Hitz pointed out in his post, there are a number of reasons why NAS might be easier and simpler to implement for virtual environments than SANs, particularly when it comes to managing .vmdk files.
But one item that hasn't been widely highlighted, and might deserve some attention, is the underlying attention required to managing LUNs in a virtualized environment. When configuring groups of virtual servers, administrators often assign some groups to one LUN set for higher performance, and other groups that are not as I/O intensive to LUNs with less performance (and more efficient costs). While helpful tools such as VMotion can easily distribute the load of virtual machines across individual servers, the .vmdk files of those virtual machines still reside on the same LUNs. So in the case of an I/O constraint causing performance issues for virtual machines, migrating that virtual machine with VMotion is not likely to solve the problem.
Configuring, assigning, and managing LUNs for virtual machines is still part art and part science. Regardless of that combination, it ultimately means more time, thinking, and management overhead for administrators. And while assigning those .vmdk files to a large NAS unit might seem like an easier method, performance concerns still present a perceived hurdle that makes many shy away from such an approach.
But alternatives exist to make use of NAS and add significant performance with a scalable caching appliance from Gear6. In these configurations, administrators can maintain one or more NAS file systems and complement those existing devices with a CACHEfx appliance to deliver high IOPS, low latency, and massive throughput. The appliance caches frequently requested data, in this case all or part of the .vmdk files, to dramatically accelerate I/O performance. Administrators can also add acceleration to a global namespace, giving them access to an almost unlimited amount of capacity without sacrificing performance. Today, it still isn't possible to dynamically resize LUNs with VMFS, so in cases where .vmdk movement for LUN expansion is a big part of the management activities, this could be a huge time savings.
There's no question that will the proliferation of dense, multi-core CPUs and multi CPU server, virtualization will play an increasing role in data center architectures. We just need to make sure that we can continue to feed those virtual machines!