SAN FRANCISCO--VMware has some new ideas for tackling an age-old problem for system administrators: how do you keep a computing service available when the server it's running on fails?
Technology called virtualization is transforming servers, higher-end systems that run ever-more-essential tasks such as e-commerce and credit card transactions. VMware, whose virtualization leadership has given the EMC subsidiary burgeoning revenue and a roaring initial public offering, naturally sees virtualization as a fix for this so-called high-availability problem.
The company's chief scientist and co-founder, Mendel Rosenblum, demonstrated two servers running e-mail software in lockstep in a speech at VMware's VMworld conference here Thursday. Through a new twist on VMware's virtualization technology, he unplugged the primary machine, and the second took over exactly where the first left off, after a couple of heartbeats' delay.
VMware co-founder and chief scientist
Mendel Rosenblum speaks Thursday
at VMworld in San Francisco.
Specialized hardware and software provides high availability today, but Rosenblum said virtualization promises to make the technology more ordinary. "The cool thing (is) that this works with any workload," he said.
The high-availability technology is an extension of the "replay" feature that exists in experimental form in VMware's current Workstation product, geared for programmers. Replay lets VMware's software log the actions of software running in a compartment called a virtual machine.
In Thursday's demonstration, the second server was executing that record of instructions performed by the first machine, taking the reins only when the first expired.
Rosenblum, who also is a Stanford University computer scientist, was careful not to promise if or when the technology might be available to users. But it's clear that the company is heading in that direction, a significant shift as rivals such as Microsoft, XenSource, Virtual Iron and Parallels are still working to get the basic virtualization technology in place.
With virtualization, software runs on a virtual layer called a hypervisor rather than on the traditional foundation, a physical machine. That separation lets many operating systems be shoehorned onto a single server for greater efficiency, and it increasingly also permits higher-level tasks, such as shifting virtual machines off of overtaxed physical machines.
VMware's primary business initially was just selling the hypervisor, but more than 80 percent of its revenue now comes from higher-level features, President Diane Greene said Tuesday.
"We're moving toward this idea of a data center that really manages the hardware itself."
--Mendel Rosenblum, chief scientist and co-founder, VMware
But it's a long way from a demo to real-world use, especially for the demanding and essential applications for which companies need high availability.
"It's a pretty serious business for the people using it," Illuminata analyst Gordon Haff said. "It remains to be seen whether it's a market (in which) VMware will be willing and interested to make the investments to really make this into a rock-solid production product."
Ultimately, virtualization will bring about a vision that server makers years ago presented--a dynamically adjusting, self-managing data center, Rosenblum said.
"What we're effectively doing is taking things that were statically assigned in the past and turning them over to a piece of software that makes decisions about how to schedule it," Rosenblum said. "We're moving toward this idea of a data center that really manages the hardware itself."
Moving a live database's data
During his speech, Rosenblum also dangled a high-availability feature for virtual machines connected to storage systems. Running virtual machines today can be moved from one physical system to another through a technology called VMotion, but Rosenblum showed a related idea he called Storage VMotion. In the demo, an administrator identified the data store of an Oracle database and moved it from one physical storage system to another, as it was in use.
Such changes are possible today, but a virtual machine must be shut down while the data store is moved.
One audience member who was keen on the Storage VMotion was Kyle Meyer, a senior systems engineer at Johnson Controls, which has 350 to 400 VMware-powered virtual machines out of a collection of 2,300 total servers. Currently, when the company's EMC Clariion storage systems go off lease, the company has to shut down servers to switch to new storage systems--a task they typically schedule at 3 a.m. Saturday in an effort to minimize disruption.
"Nondisruption is the biggest thing," Meyer said. Today, "we lease Clariion. If we're moving it to somebody else, we've got to shut it down."
There are complications to using Storage VMotion, Rosenblum said in an interview after the speech. For one thing, moving a data store takes a long time, even with the "teeny" one and the high-speed network in the demo. For another, a database that's already running full-tilt might not have network capacity to spare for being moved to another storage system.
"If your workload is maxing out, anything you do to copy (the data store) would show some degradation," he said.
A third demonstration showed a faster way of downloading virtual machines. In VMware's vision, computer users will commonly fire up prebuilt virtual machines for particular tasks such as secure Web browsing, but the size of virtual machines makes fetching such "virtual appliances" impractical.
In the demonstration, the VMware instant-on technology fetches the most important virtual-appliance data first, so it can begin its virtual-boot process while more data streams in. It took about 2 minutes to boot a 410MB virtual appliance, a version of Ubuntu Linux tailored for Web browsing, into a usable state.
"We see this as an incredibly exciting way of packaging software that gets rid of a lot of problems," Rosenblum said of the appliance approach.