VMware Inc. last week launched the long-awaited upgrade of its core virtualization software, which it calls a cloud operating system.
First touted by CEO Paul Maritz during VMware's user conference last fall, the new vSphere 4 software can transform data centers into “virtual compute clouds” in which applications move fluidly across computing, network and storage resources, company executives contended last week.
The hope, they said during a press conference at VMware's Palo Alto, Calif., headquarters, is that the upgrade will convince IT executives that virtualization is now reliable and scalable enough to run large corporate databases and other critical applications.
At the heart of VMware's promise to extend cloud computing to the data center are its claims that the software can now manage a cluster of computers with up to 32 physical servers, 2,048 processor cores and 32TB of memory. Perhaps more important for database applications, VMware said it has doubled the maximum number of I/O operations its software can perform to more than 200,000 per second.
Beyond the hype and bluster from executives at last week's announcement, some analysts and customers said that although some of VMware's claims are valid, the software still lacks some key features.
Chris Wolf, an analyst at Burton Group in Midvale, Utah, said that new processor technologies, combined with improvements in the virtualization software, will allow vSphere to run more-demanding applications.
However, Wolf added that VMware still doesn't offer tools that can provide all the management capabilities needed to treat a cluster of industry-standard servers as if it were a mainframe computer or a large Unix system. For example, VMware DRS, a tool that monitors processor and memory usage in a pool of servers to determine the most efficient platform available to run a virtual machine, still doesn't measure some key factors, like I/O utilization, he said.
And while Maritz last week talked about the danger of “lock-in” with proprietary systems from other vendors, vSphere still can't manage hypervisors from companies like Microsoft Corp. and Citrix Systems Inc. VMware executives argued that there isn't enough demand for those vendors' offerings to justify the investment required to support them, but some customers said they could use those capabilities today.
Christopher Rence, CIO at Fair Isaac Corp. (Fico), a Minneapolis-based provider of decision management tools and services, said his company has mostly standardized on VMware products but is also piloting Microsoft's Hyper-V virtualization technology and plans to use it for some applications. He noted that Fico must support the Microsoft offering because some of its customers run Hyper-V in their data centers. Thus, the ability to manage Hyper-V with vSphere would benefit the company.
Nonetheless, Rence said that VMware's technology has so far helped Fico consolidate 24 data centers into four. The company uses VMware widely in its development and testing processes and is now piloting some large database projects on vSphere, including applications that process up to 1,100 transactions per second.
Fico also uses a new technology in vSphere called vShield Zones to enforce security and compliance policies across a group of virtual machines. The technology makes it possible for virtual machines to maintain their security policies even if Fico moves them to a different cluster pool.
Other new features in vSphere 4 include fault-tolerance capabilities, support for thin storage provisioning — which allows less physical storage to be allocated to a virtual machine — the ability to switch an application between storage systems while the application is running, and a “distributed switch” developed with Cisco Systems Inc.
VSphere 4 is slated to be available later in the current quarter. Pricing ranges from $995 for a three-server package to $3,495 per processor for a high-end package called vSphere 4 Enterprise Plus.
Maritz said the low-end offering allows small companies to virtualize a pool of three or four servers so that a workload can fail over to a different system in the event of a hardware failure.