Speaking on the third and final day of VMworld 2007, VMware's annual conference, Rosenblum told a rapt audience that virtualization is not "some magic pixie dust that you sprinkle over the data center that makes everything run better." Instead, he showed how VMware, by applying what he called "a layer of indirection" between an operating system and hardware, will continue to solve "some of computer science's hardest problems."
VMotion for storage
One of those persistent problems is downtime. For example, Rosenblum showed how future products will leverage what he dubbed "VMotion for disks" to move not just virtual machines (VMs) but also their underlying virtual disks to different physical storage platforms while the virtual machines are still running.
"Why is this interesting?" Rosenblum asked. For one thing, customers tend to lease storage, and when the lease comes up, swapping out an old for a new array "tends to involve shutting down the VMs," he said. With VMotion for disks, virtual disks could be moved in the background, without incurring any downtime. Similarly, VMotion for disks could be used to optimize virtual disk placement on a higher-performance array or a less utilized one. VMotion for disks will be storage agnostic; virtual disks migration will be possible across entirely different storage architectures, such as Fibre Channel storage area networks (SANs), iSCSI SANs, or even NAS arrays."You can do all this because you have this level of indirection," he said.
Streaming virtual machines
Another compelling demo showed how VMware is working to improve its virtual appliance offering, which may also be applied to disaster recovery and virtual desktop infrastructure (VDI).
Fault-tolerant virtual machines
But the audience reserved its most thunderous applause for a feature Rosenblum called Continuous Availability, which would give ordinary virtual machines levels of fault tolerance that have traditionally been reserved for only the most expensive server systems.
Continuous Availability would build on that idea, except that instead of storing the executions in a file, it would send them over to a secondary VM on a separate host. Then, in the event that the host of the primary VM went "poof," as Rosenblum put it, the second VM is ready to take over, and "nothing is lost."Rosenblum and his assistant demonstrated Continuous Availability on two physical servers running Microsoft Exchange VMs in lockstep. Rosenblum pulled both plugs from the primary server, and within seconds the secondary VM had taken over without being restarted.
Scratching the surface
VMware won't stop there, Rosenblum said. "We've only just begun to benefit from virtualization," and since it's "pretty clear that this is a better way of doing things, there will be very rapid adoption."
Let us know what you think about the story; email Alex Barrett, News Director..