It all started with a pretty run-of-the-mill partnership agreement. FalconStor announced its storage virtualization, snapshot and replication software will support Virtual Iron’s virtual servers last week, and my SearchServerVirtualization.com colleague Alex Barrett and I agreed to take briefings.
FalconStor is walking a s tightrope here, because it’s also partnered with Virtual Iron’s large rival VMware. But FalconStor has to come up with reasons to use Virtual Iron over VMware, (i.e. ways to promote that partnership). This led Alex to begin an interesting conversation with FalconStor’s vice president of business development Bernie Wu about the pros and cons of virtualizing storage with VMware vs. Virtual Iron. Wu pointed out what he’d later reprise in a separate call with me: that the use case for FalconStor’s IPStor storage virtualization software is in many ways stronger with Virtual Iron, because VI doesn’t have its own file system, like VMware does.
As Burton Group senior analyst Chris Wolf patiently explained to me later, VMware’s file system means that its hypervisor (the software layer that controls the host server, guest OSes, and their interaction with the rest of the network) is handling virtual hard disk mapping on back-end storage systems. You can use VMware with raw device mapping (RDM), but then you turn off many of the features VMware users have come to like about VMware, such as VMotion. (RDM also has a slightly limited “virtual mode” as of 3.0, but that’s a tangential discussion.) This makes virtual hard disk mapping performed by storage virtualization products, whether appliances or software, at least somewhat redundant.
So I asked Wu, “what are users missing out on if they can’t use your storage virtualization software with VMware?” His first answer was large-scale data migrations.
Up until VMware’s Virtual Infrastructure 3.5, VMware had no ability to move the data it managed in its virtual hard disks on back-end storage; hence storage virtualization devices stepped in to fill the gap. With Storage VMotion in 3.5, that gap was at least partially closed. Storage VMotion is still a difficult way to do a large-scale migration, however, because it migrates data one host at a time. So storage virtualization devices, which perform migrations en masse, still have that advantage. At least, until and unless Storage VMotion adds that capability.
Aside from large-scale migrations, Wu also told me that thin provisioning is another capability IPStor offers that VMware doesn’t. That’s a big deal–VMware’s best practices recommend that users allot twice the amount of disk space they actually plan to write to; the ability to expand capacity on the fly helps everyone avoid buying 2x the amount of storage they need.
The Burton Group’s Wolf pointed out plenty more gaps in VMware’s storage capabilities–heterogeneous array support; heterogeneous multiprotocol support (Storage VMotion doesn’t support iSCSI yet); I/O caching; and heterogeneous replication support.
Some of these gaps will likely be filled by VMware or the storage industry. For instance, when it comes to multiprotocol support, VMware’s MO with new features has always been to support Fibre Channel first and they usually get around to iSCSI soon after. And what happens to the need for heterogeneous multiprotocol support if FCoE ever takes off? What of I/O caching, when and if everybody’s working with 10 Gigabit Ethernet pipes? And VMware’s launching its own management software for heterogeneous replication support (even if it’s not doing the replication itself).
So it seems that storage virtualization players will have to start coming up with more value-adds for VMware environments as time goes on.
VMware has its own tightrope to walk, too. Take replication for example–VMware supports replication from its partners, saying it doesn’t want to reinvent the wheel. But that’s the kind of thing it said when users were asking for Storage VMotion back in 2006, too.
“Deep down, I believe that VMware isn’t going to push its partners out,” Wolf said. And indeed, VMware did make a good-faith gesture last fall with the announcement of a certification program for storage virtualization partners. Wolf also pointed out, “A lot of organizations are afraid to deploy something in the data path unless their hardware vendors will support and certify it–without the support of their partners, VMware would have a tough time playing ball.”
But that might not be the case as much now as back when EMC first bought VMware in 2003 and everybody in the storage world scrateched their heads and wondered why. Now, VMware has its own muscles to flex as its billion-dollar 2007 IPO for 10 percent of the company proved.
More and more analysts are telling me that the hypervisor will become the data center operating system of the future. Over in the server virtualization world, Wolf says VMware competitors argue that the hypervisor is a commodity, and VMware says it isn’t. “In order to keep the hypervisor from becoming commoditized, they have to keep adding new features,” he said.
Which suggests to me that storage virtualization vendors should probably be working on new features, too.