A virtualised server that experiences a catostrophic failure has the potential to knock offline up to 20 virtual machines, and hence is driving the need for ultra reliable servers. So says Stratus Technologies, which has just doubled the performance of its new ftServer line thanks to the use of Intel's high-performance Nehalem microarchitecture.

The new ftServer line is in fact the fifth-generation of the line of servers designed to provide businesses with a completely fault tolerant server that can provide continuous or high availability for mission-critical Windows applications.

The ftServer 2600, 4500 and 6300 models (based on Intel Nehalem processors) are one- and two-socket servers delivering quad-core processing power. Stratus is stressing the easy of use with these servers, which are designed to used straight out of the box.

The company says that applications can run with no software modification or failover scripting required, and immediately benefit from the processing power of all eight logical cores and nearly 99.9999 percent availability.

"The unique factor about ftServers is that double splices fit into a high speed back-plain, so that two independent units are presented as a single machine to the outside world," Andy Bailey, Availability Consultant for Stratus. "This is not a complex cluster arrangement, but rather one system that can be managed as an Intel-based server."

"We adhere to a policy to keep everything as simple as possible from the user perspective," he told Techworld. To that end, most components are hot swappable and are customer replaceable. "It offers zero latency," Bailey said. "There is no concept of secondary or primary, it is not mirroring, it is just doing things at the same time with synchronisation maintained down to the millisecond."

But what exactly does the new ftServer systems offer compared to previous models? Well for one they sport the first fault-tolerant implementation of Intel Xeon 5500 processors, which features multi-core technology that intelligently maximises performance to match workload. It also contains the first fault-tolerant implementation of QuickPath Interconnect Technology (QPI) for dynamically scalable interconnect bandwidth and memory performance to address a wider spectrum of database requirements.

Memory has been tripled up to 96GB and now eight pair of 2.5 inch internal SAS disks offer up to 8TB of physical storage. There is now also an integrated virtual technician module for proactive fault notification, remote system access, and improved online upgrading.

"We have always used Intel processors," said Bailey. "The whole Nehalem architecture offers a significant boosts across the ftServer systems, although processor speed on paper is not higher (than previous models), the architecture changes means faster processing around memory access, updated PCI Express 2 spec, 2.5 in drives, greater capacity and spindles, which has increased the I/O rates."

But what is driving the need for high availability systems? "Virtualisation really is a driver for uptime systems," said Bailey. "A lot of customers address virtualisation through software. There are good products out there, but the fundamental problem with any HA system is that you have to restart it after a catostrophic failure. With virtualisation, if one platform fails, it can take down 20 virtual machines running on that machine. We are seeing increased interest because of that."

The entry-level ftServer system (2600) costs in the region of £12,000 ($19,700) to £14,000 ($22,988), whereas the mid range 4500 system comes in at between £22,000 to £23,000. The high-end 6300, costs in the region of £30,000 to £35,000 upwards.

Follow Tom Jowitt on Twitter: http://twitter.com/TJowitt