Marathon Technologies is adding fault tolerance -- the ability to carry on operations without data loss in the face of a system failure -- on top of Citrix XenServer virtual machines.

If virtualized fault tolerance becomes readily available, it could shift the nature of the workloads that virtualized servers are able to carry from secondary to mission-critical applications.

Marathon is the supplier of everRun, a software-based fault-tolerance system for hardware servers that's been deployed by 1,800 organizations since it was introduced in 2004, according to CEO Gary Phillips. Fault tolerance was pioneered by Tandem Computers in the late 1970s, when its NonStop hardware systems showed they could override any single hardware or software component failure. Vendors already offer rapid failover and recovery of virtual machines, but fault tolerance with no interruption or lost data is an advance in the state of the art for virtual machine operations.

Phillips said in an interview that Marathon's everRun has been adapted to meet the needs of virtualized servers. On Monday, he said everRun VM will become available at the end of April to provide fault tolerance for Citrix Systems' XenServer, a hypervisor based on open source Xen. Marathon's partnership with Citrix means it's focused on providing fault tolerance to virtual machines running Windows applications in a Windows-based virtual machine only.

Citrix senior VP Peter Levine hailed the Marathon move as an extension of XenServer's capabilities. EverRun has been "seamlessly integrated" with XenServer, he said. "The combination allows companies to virtualize a range of applications that they wouldn't even consider previously," he said in a statement accompanying Marathon's announcement.

Marathon CTO Jerry Melnick cited the following features in everRun VM:

Existing Windows applications get assurance of continuous operation without needing to be modified, Melnick said.

Fault tolerance has typically been a complex, labor-intensive feature to add to a server or cluster of servers. "We've gone to great lengths to automate availability. We are taking out labor and overhead," Melnick said. EverRun sits between the hypervisor and the virtual machine, monitoring the traffic between them and prepared to reroute that traffic.

EverRun can supply simple failover for an application. At a higher level, it can protect against hardware component or complete hardware system failure, with a mirrored system ready to go in the event of an outage. It can also protect against data center disaster by placing the failover virtual machines and hardware at a different geographic location.

When asked if virtual machines running Linux will be added to the mix, Phillips said, "Not this year." But Marathon has plans for a development project next year to do so; no date of delivery is available at this time. EverRun VM will be priced at $4,500 and will include the XenServer hypervisor; if a customer owns XenServer, everRun can be added for $2,000 per hypervisor.

High-availability or fault-tolerant operations are estimated to cover less than 20% of servers currently running, according to research figures cited by Marathon.