Stratus said its fault-tolerant systems running VMware can supply 99.9999% continuous operation, or the equivalent of 32 seconds of unplanned hardware outage a year. The more common four nines (99.99%) allows 52 minutes of downtime a year; five nines equals five minutes of downtime.
Stratus servers consist of two systems with separate CPUs, network interface cards, and other components sharing a common backplane. Both sets of components run all application instructions simultaneously. If a hardware component fails, the other live set of components picks up the processing load automatically, without interruption.
"There's no failover [because] the two things are running in complete lockstep. They're so tightly linked the operating system doesn't know there's two sets of hardware," said Denny Lane, director of product marketing, in an interview.
Databases running online transaction processing and other mission-critical applications have been holdouts in the general wave of virtualizing the data center. Their workloads have been considered too valuable to risk taking off their standalone servers and running in virtualized environments to obtain higher hardware utilization rates.
Stratus builds servers starting at $10,000 using Intel Xeon chips to supply a low-cost form of fault tolerance on the pattern set by Tandem Computer's NonStop systems. It also produces systems to run HP-UX Unix. It has 11,000 servers deployed around the world, Lane said.
Stratus servers conduct their own continuous error checking as a feature of the virtual operating system. With the base or Foundation version of VMware Infrastructure 3 bundled in, a Stratus server can run Windows Server and Red Hat Enterprise Linux applications in virtual machines side by side without fear of incompatibilities or interruption, Lane said.
He said the addition of a virtual machine system on the servers means more and more enterprise workloads can take on a continuous availability characteristic, and the server can still be heavily utilized.
High availability also can be achieved through virtualized backups, but they do involve a software failover from one physical server to another, a few seconds of delay, and a server restart on a clustered machine, Lane said.
InformationWeek Analytics has published an independent analysis of the challenges around virtualization management. Download the report here (registration required).